From e1dcd386800ebdad036ecbff573cae6e582c79ad Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 20 Oct 2023 12:46:30 +0800 Subject: [PATCH 01/56] fix: code interpreter program join --- creator/code_interpreter/R.py | 6 +++--- creator/code_interpreter/base.py | 20 +++++++++++++++----- 2 files changed, 18 insertions(+), 8 deletions(-) diff --git a/creator/code_interpreter/R.py b/creator/code_interpreter/R.py index 6087685..f6e083a 100644 --- a/creator/code_interpreter/R.py +++ b/creator/code_interpreter/R.py @@ -6,14 +6,14 @@ class RInterpreter(BaseInterpreter): name: str = "r_interpreter" description: str = "An R interpreter" start_command: str = "R --quiet --no-save --no-restore-data" - print_command: str = "cat('{}\n')" + print_command: str = "'{}'\n" def postprocess(self, response): def clean_string(s): return '\n'.join([line for line in s.split('\n') if not re.match(r'^(\s*>\s*|\s*\.\.\.\s*)', line)]) - + # clean up stdout and stderr response['stdout'] = clean_string(response.get('stdout', '')) response['stderr'] = clean_string(response.get('stderr', '')) - + return response diff --git a/creator/code_interpreter/base.py b/creator/code_interpreter/base.py index 236864f..c10d3c8 100644 --- a/creator/code_interpreter/base.py +++ b/creator/code_interpreter/base.py @@ -41,11 +41,12 @@ def handle_stream_output(self, stream, is_stderr): """Reads from a stream and appends data to either stdout_data or stderr_data.""" start_time = time.time() for line in stream: + if self.done.is_set(): + break if self.detect_program_end(line): - start_time = time.time() + self.done.set() break if time.time() - start_time > self.timeout: - start_time = time.time() self.output_cache["stderr"] += f"\nsession timeout ({self.timeout}) s\n" break if line: @@ -68,6 +69,15 @@ def clear(self): self.stdout_thread.start() self.stderr_thread.start() + def join(self, timeout): + start_time = time.time() + + while time.time() - start_time < timeout: + if self.done.is_set(): + break + self.stdout_thread.join(0.1) + self.stderr_thread.join(0.1) + def preprocess(self, code): return code @@ -92,7 +102,7 @@ def run(self, query: str, is_start: bool = False) -> dict: query = self.add_program_end_detector(query) self.process.stdin.write(query + "\n") self.process.stdin.flush() - + time.sleep(0.2) except subprocess.TimeoutExpired: self.process.kill() @@ -101,7 +111,7 @@ def run(self, query: str, is_start: bool = False) -> dict: except BrokenPipeError: stderr = traceback.format_exc() return {"status": "error", "stdout": "", "stderr": stderr} - + self.join(self.timeout) return self.postprocess({"status": "success", **self.output_cache}) def __del__(self): @@ -110,4 +120,4 @@ def __del__(self): if self.stdout_thread: self.stdout_thread.terminate() if self.stderr_thread: - self.stderr_thread.terminate() \ No newline at end of file + self.stderr_thread.terminate() From 18cef46c85507c1efa82ea776184e51a2bceb4d2 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 20 Oct 2023 13:03:51 +0800 Subject: [PATCH 02/56] fix: skill install raise exception --- creator/core/skill.py | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-) diff --git a/creator/core/skill.py b/creator/core/skill.py index 130a0b7..ff54463 100644 --- a/creator/core/skill.py +++ b/creator/core/skill.py @@ -199,15 +199,20 @@ def to_skill_function_schema(self): return code_skill_json_schema - def check_and_install_dependencies(self): + def install(self): if self.skill_dependencies is None: return - install_script = generate_install_command(self.skill_program_language, self.skill_dependencies) - result = config.code_interpreter.run({ - "language": "shell", - "code": install_script, - }) - print(result, print_type="json") + try: + install_script = generate_install_command(self.skill_program_language, self.skill_dependencies) + print("> Installing dependencies", print_type="markdown") + print(f"```bash\n{install_script}\n```\n", print_type="markdown") + result = config.code_interpreter.run({ + "language": "shell", + "code": install_script, + }) + print(f"> Install dependencies result: {result}", print_type="markdown") + except Exception as e: + print(f"> Error when installing dependencies: {e}", print_type="markdown") return def __add__(self, other_skill): @@ -242,7 +247,7 @@ def __gt__(self, user_request:str): return self.refactor() def run(self, inputs: Union[str, dict[str, Any]]): - self.check_and_install_dependencies() + self.install() messages = [ {"role": "assistant", "content": "ok I will run your code", "function_call": { "name": self.skill_name, @@ -277,7 +282,7 @@ def test(self): previews_tool = code_tester_agent.tools[0] code_tester_agent.tools[0] = config.code_interpreter - self.check_and_install_dependencies() + self.install() extra_import = """\n\n import io import unittest From 457d7f668a5704060889365a5d8d8546a82d0e29 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 20 Oct 2023 13:19:42 +0800 Subject: [PATCH 03/56] fix: skill_dependencies parse --- creator/core/skill.py | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-) diff --git a/creator/core/skill.py b/creator/core/skill.py index ff54463..64b1823 100644 --- a/creator/core/skill.py +++ b/creator/core/skill.py @@ -152,15 +152,28 @@ def __init__(self, **data): if isinstance(data["skill_parameters"], list): self.skill_parameters = [CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**param)) for param in data["skill_parameters"]] elif isinstance(data["skill_parameters"], dict): - self.skill_parameters = CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**data["skill_parameters"])) + if self.skill_parameters.param_name in ("null", "None") or self.skill_parameters.param_type in ("null", "None"): + self.skill_parameters = [] + else: + self.skill_parameters = [CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**data["skill_parameters"]))] if "skill_return" in data and data["skill_return"]: if isinstance(data["skill_return"], list): self.skill_return = [CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**param)) for param in data["skill_return"]] elif isinstance(data["skill_return"], dict): - self.skill_return = CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**data["skill_return"])) if self.skill_return.param_name in ("null", "None") or self.skill_return.param_type in ("null", "None"): - self.skill_return = None + self.skill_return = [] + else: + self.skill_return = [CodeSkillParameter(**CodeSkillParameter.construct_with_aliases(**data["skill_return"]))] + + if "skill_dependencies" in data and data["skill_dependencies"] is not None: + if isinstance(data["skill_dependencies"], list): + self.skill_dependencies = [CodeSkillDependency(**dependency) for dependency in data["skill_dependencies"]] + elif isinstance(data["skill_dependencies"], dict): + if self.skill_dependencies.dependency_name in ("null", "None") or self.skill_dependencies.dependency_name in ("null", "None"): + self.skill_dependencies = [] + else: + self.skill_dependencies = [CodeSkillDependency(**data["skill_dependencies"])] def to_function_call(self): parameters = { From 97a3eddb04ad37879829439c9e825e0018f89120 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 20 Oct 2023 13:56:43 +0800 Subject: [PATCH 04/56] fix: add set done when timeout --- creator/code_interpreter/base.py | 1 + 1 file changed, 1 insertion(+) diff --git a/creator/code_interpreter/base.py b/creator/code_interpreter/base.py index c10d3c8..ef392ef 100644 --- a/creator/code_interpreter/base.py +++ b/creator/code_interpreter/base.py @@ -47,6 +47,7 @@ def handle_stream_output(self, stream, is_stderr): self.done.set() break if time.time() - start_time > self.timeout: + self.done.set() self.output_cache["stderr"] += f"\nsession timeout ({self.timeout}) s\n" break if line: From 260f8c5cf75cc28c834196730c9cb8714b5aef72 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Sat, 21 Oct 2023 14:22:32 +0800 Subject: [PATCH 05/56] fix: handle non skill return and refact llm create name --- creator/core/core.py | 3 +++ creator/llm/{chatopenai_with_trim.py => chatllm_with_trim.py} | 0 creator/llm/llm_creator.py | 2 +- 3 files changed, 4 insertions(+), 1 deletion(-) rename creator/llm/{chatopenai_with_trim.py => chatllm_with_trim.py} (100%) diff --git a/creator/core/core.py b/creator/core/core.py index 94d5c42..9f09bd0 100644 --- a/creator/core/core.py +++ b/creator/core/core.py @@ -101,6 +101,9 @@ def _create_from_messages(cls, messages) -> CodeSkill: "messages": messages, "verbose": True, }) + if skill_json is None: + print("> No skill was generated.", print_type="markdown") + return None skill = CodeSkill(**skill_json) if skill.skill_metadata is None: skill.skill_metadata = BaseSkillMetadata() diff --git a/creator/llm/chatopenai_with_trim.py b/creator/llm/chatllm_with_trim.py similarity index 100% rename from creator/llm/chatopenai_with_trim.py rename to creator/llm/chatllm_with_trim.py diff --git a/creator/llm/llm_creator.py b/creator/llm/llm_creator.py index a3e571e..73dd829 100644 --- a/creator/llm/llm_creator.py +++ b/creator/llm/llm_creator.py @@ -2,7 +2,7 @@ from creator.callbacks import OutputBufferStreamingHandler, RichTerminalStreamingHandler, FileLoggerStreamingHandler from langchain.callbacks.manager import CallbackManager from langchain.embeddings import OpenAIEmbeddings -from .chatopenai_with_trim import ChatOpenAIWithTrim, AzureChatOpenAIWithTrim +from .chatllm_with_trim import ChatOpenAIWithTrim, AzureChatOpenAIWithTrim def create_llm(config): From 9a46ec9d7dec9d958d67e28cc7378c4049cf532a Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Sat, 21 Oct 2023 17:20:25 +0800 Subject: [PATCH 06/56] =?UTF-8?q?feat:=20add=20version=20and=20update=20cu?= =?UTF-8?q?rsor=20=E2=97=8F?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- creator/__version__.py | 2 +- creator/callbacks/rich_manager.py | 31 +++++++++++++++++-------------- pyproject.toml | 2 +- 3 files changed, 19 insertions(+), 16 deletions(-) diff --git a/creator/__version__.py b/creator/__version__.py index b3f4756..ae73625 100644 --- a/creator/__version__.py +++ b/creator/__version__.py @@ -1 +1 @@ -__version__ = "0.1.2" +__version__ = "0.1.3" diff --git a/creator/callbacks/rich_manager.py b/creator/callbacks/rich_manager.py index a561822..5318c72 100644 --- a/creator/callbacks/rich_manager.py +++ b/creator/callbacks/rich_manager.py @@ -84,19 +84,18 @@ def finish(self, message=None, err=None) -> None: def refresh_text(self, cursor: bool = True) -> None: """Refreshes the content display.""" text = self.content + replacement = "```text" lines = text.split('\n') inside_code_block = False - for line in lines: + for i, line in enumerate(lines): # find the start of the code block - if line.startswith("```"): + if line.strip().startswith("```"): inside_code_block = not inside_code_block + if inside_code_block: + lines[i] = replacement content = '\n'.join(lines) - if cursor: - content += "█" - else: - if content.endswith("█"): - content = content[:-1] + content = self.update_cursor(cursor, content) if inside_code_block: content += "\n```" markdown = Markdown(content.strip()) @@ -109,7 +108,8 @@ def refresh_code(self, cursor: bool = True) -> None: code_table = self._create_code_table(cursor) output_panel = self._create_output_panel() - group = Group(code_table, output_panel) + group_items = [code_table, output_panel] + group = Group(*group_items) self.code_live.update(group) self.code_live.refresh() @@ -122,6 +122,13 @@ def refresh(self, cursor: bool = True, is_code: bool = True) -> None: else: self.refresh_text(cursor=cursor) + def update_cursor(self, cursor, content): + if cursor: + content += "●" + elif content.endswith("●"): + content = content[:-1] + return content + def update_tool_result(self, chunk): if not self.use_rich: return @@ -167,11 +174,7 @@ def _create_code_table(self, cursor: bool) -> Panel: code_table = Table(show_header=False, show_footer=False, box=None, padding=0, expand=True) code_table.add_column() - if cursor: - self.code += "█" - else: - if len(self.code) > 0 and self.code[-1] == "█": - self.code = self.code[:-1] + self.code = self.update_cursor(cursor, self.code) code_lines = self.code.strip().split('\n') for i, line in enumerate(code_lines, start=1): @@ -193,7 +196,7 @@ def _get_line_syntax(self, line: str, line_number: int) -> Syntax: def _create_output_panel(self) -> Panel: """Creates a panel for displaying the output.""" if not self.tool_result or self.tool_result == "None": - return Panel("", box=MINIMAL, style="#FFFFFF on #3b3b37") + return "" return Panel(self.tool_result, box=MINIMAL, style="#FFFFFF on #3b3b37") diff --git a/pyproject.toml b/pyproject.toml index 289108f..8987f03 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -3,7 +3,7 @@ name = "open-creator" packages = [ {include = "creator"}, ] -version = "0.1.2" +version = "0.1.3" description = "Build your costomized skill library" authors = ["JunminGONG "] readme = "README.md" From 0ce97ed2df7de3cb9c76143e1a58b46520da7b54 Mon Sep 17 00:00:00 2001 From: Sayo Date: Sat, 21 Oct 2023 18:00:36 +0800 Subject: [PATCH 07/56] [feat] add expertise_prompt_agent --- creator/agents/expertise_prompt_agent.py | 84 +++++++++++++++++++ creator/config/library.py | 1 + .../prompts/expertise_prompt_agent_prompt.md | 15 ++++ 3 files changed, 100 insertions(+) create mode 100644 creator/agents/expertise_prompt_agent.py create mode 100644 creator/prompts/expertise_prompt_agent_prompt.md diff --git a/creator/agents/expertise_prompt_agent.py b/creator/agents/expertise_prompt_agent.py new file mode 100644 index 0000000..a4e8904 --- /dev/null +++ b/creator/agents/expertise_prompt_agent.py @@ -0,0 +1,84 @@ +from typing import Any, Dict + +from langchain.prompts import ChatPromptTemplate +from langchain.output_parsers.json import parse_partial_json + +from creator.config.library import config +from creator.utils import load_system_prompt +from creator.llm.llm_creator import create_llm + +from base import BaseAgent + + +class ExpertisePromptAgent(BaseAgent): + output_key: str = "expertise_prompt" + + @property + def _chain_type(self): + return "ExpertisePromptAgent" + + def construct_prompt(self, langchain_messages: Dict[str, Any]): + prompt = ChatPromptTemplate.from_messages(messages=[ + *langchain_messages, + ("system", self.system_template) + ]) + return prompt + + def parse_output(self, messages): + function_call = messages[-1].get("function_call", None) + if function_call is not None: + rewrited_prompt = parse_partial_json(function_call.get("arguments", "{}")) + prefix_prompt = rewrited_prompt.get("prefix_prompt", "") + postfix_prompt = rewrited_prompt.get("postfix_prompt", "") + result = {"prefix_prompt": prefix_prompt, "postfix_prompt": postfix_prompt} + return {"expertise_prompt": result} + return {"expertise_prompt": None} + + +def create_expertise_prompt_agent(llm): + template = load_system_prompt(config.expertise_prompt_agent_prompt_path) + + function_schema = { + "name": "expertise_prompt", + "description": "a function that guide GPT to act as an expert in the field to respond to the user's request", + "parameters": { + "properties": { + "prefix_prompt": { + "type": "string", + "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset.", + }, + "postfix_prompt": { + "type": "string", + "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response.", + } + }, + "type": "object", + "required": ["prefix_prompt", "postfix_prompt"] + } + } + + chain = ExpertisePromptAgent( + llm=llm, + system_template=template, + function_schemas=[function_schema], + verbose=False + ) + return chain + + +llm = create_llm(config) +expertise_prompt_agent = create_expertise_prompt_agent(llm) + +if __name__ == "__main__": + TabMWP_test = """\"question\": \"Hannah baked cookies each day for a bake sale. How many more cookies did Hannah bake on Saturday than on Sunday?\", \"table\": \"Name: Cookies baked\nUnit: cookies\nContent:\nDay | Number of cookies\nFriday | 163\nSaturday | 281\nSunday | 263\" +""" + tool_transfer_test=""""intro": "The tool is used to calculate the optimal number of units to produce to maximize profit for a manufacturing company. It takes into account the fixed costs, variable costs, selling price, and demand for the product. The function uses the formula Profit = (Selling Price * Quantity) - (Variable Cost * Quantity) - Fixed Cost to calculate the profit and returns the optimal quantity to produce.", "tool": "```python\ndef calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand):\n \"\"\"\n Calculates the optimal number of units to produce to maximize profit.\n\n Parameters:\n selling_price (float): The price at which the product is sold.\n variable_cost (float): The cost of producing one unit of the product.\n fixed_cost (float): The fixed cost of production.\n demand (int): The number of units that can be sold at the given price.\n\n Returns:\n int: The optimal number of units to produce to maximize profit.\n \"\"\"\n # Calculate the profit for each quantity\n profits = []\n for quantity in range(1, demand+1):\n profit = (selling_price * quantity) - (variable_cost * quantity) - fixed_cost\n profits.append(profit)\n\n # Find the quantity that maximizes profit\n optimal_quantity = profits.index(max(profits)) + 1\n\n # Return the optimal quantity\n return optimal_quantity\n```", "scn1": "Production Planning\nA manufacturing company produces a product that has a fixed cost of $10,000, a variable cost of $5 per unit, and a selling price of $20 per unit. The company can sell up to 5,000 units of the product at this price. What is the optimal number of units to produce to maximize profit?", "sol1": "```python\n# Set the inputs for the manufacturing company\nselling_price = 20\nvariable_cost = 5\nfixed_cost = 10000\ndemand = 5000\n\n# Calculate the optimal number of units to produce\noptimal_quantity = calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand)\n\n# Print the optimal quantity\nprint(optimal_quantity)\n```", "ans1": 5000.0, "scn2": "Pricing Strategy\nA company produces a product that has a fixed cost of $20,000, a variable cost of $10 per unit, and a demand of 10,000 units. The company wants to maximize profit and is considering two pricing strategies. The first strategy is to sell the product at $30 per unit, and the second strategy is to sell the product at $35 per unit. What is the optimal pricing strategy for the company?", "sol2": "```python\n# Set the inputs for the company\nfixed_cost = 20000\nvariable_cost = 10\ndemand = 10000\n\n# Strategy 1: Selling the product at $30 per unit\nselling_price_1 = 30\noptimal_quantity_1 = calculate_optimal_units(selling_price_1, variable_cost, fixed_cost, demand)\nprofit_1 = (selling_price_1 * optimal_quantity_1) - (variable_cost * optimal_quantity_1) - fixed_cost\n\n# Strategy 2: Selling the product at $35 per unit\nselling_price_2 = 35\noptimal_quantity_2 = calculate_optimal_units(selling_price_2, variable_cost, fixed_cost, demand)\nprofit_2 = (selling_price_2 * optimal_quantity_2) - (variable_cost * optimal_quantity_2) - fixed_cost\n\n# Determine the optimal pricing strategy\nif profit_1 > profit_2:\n print(\"The optimal pricing strategy is to sell the product at $30 per unit.\")\nelse:\n print(\"The optimal pricing strategy is to sell the product at $35 per unit.\")\n```", "ans2": 35.0, "scn3": "Capacity Planning\nA company produces a product that has a fixed cost of $50,000, a variable cost of $15 per unit, and a selling price of $25 per unit. The company has a production capacity of 10,000 units. What is the optimal number of units to produce to maximize profit?", "sol3": "```python\n# Set the inputs for the company\nselling_price = 25\nvariable_cost = 15\nfixed_cost = 50000\ndemand = 10000\n\n# Calculate the optimal number of units to produce\noptimal_quantity = calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand)\n\n# Print the optimal quantity\nprint(optimal_quantity)\n```" +""" + messages = [{ + "role": "user", + "content": TabMWP_test + }] + result = expertise_prompt_agent.run({ + "messages": messages, + "verbose": True, + }) diff --git a/creator/config/library.py b/creator/config/library.py index a033469..9b53611 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -86,6 +86,7 @@ class LibraryConfig(BaseModel): tips_for_debugging_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_debugging_prompt.md") tips_for_testing_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_testing_prompt.md") tips_for_veryfy_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_veryfy_prompt.md") + expertise_prompt_agent_prompt_path: str = os.path.join(project_dir, "prompts", "expertise_prompt_agent_prompt.md") use_rich: bool = True use_file_logger: bool = False diff --git a/creator/prompts/expertise_prompt_agent_prompt.md b/creator/prompts/expertise_prompt_agent_prompt.md new file mode 100644 index 0000000..75fa62f --- /dev/null +++ b/creator/prompts/expertise_prompt_agent_prompt.md @@ -0,0 +1,15 @@ +You are a "Prompt Advisor". Your task is to guide GPT to respond as an expert in a given field based on the user's request. Provide a short prefix_prompt as specific instructions for the expert role. Then, offer brief tips in a postfix_prompt. The user's original request will be placed between the two prompts. + +Here is an example: +### +User_Request: +``` +A sphere is inscribed in a cone with height 4 and base radius 3. What is the ratio of the volume of the sphere to the volume of the cone?\n\n[asy]\nfill(circle((1.5,0),1.5),gray(.7));\ndraw((0,3)--(4,0)--(0,-3));\ndraw(circle((1.5,0),1.5));\ndraw((0,0)..(1.5,-.3)..(3,0));\ndraw((0,0)..(1.5,.3)..(3,0),dashed);\ndraw(xscale(.15)*circle((0,0),3));\n[/asy] +``` + +result: +``` +"prefix_prompt": "You are doing a math test as a mathematically-trained expert, well-versed in a wide range of mathematical concepts and terms; your main objective is to comprehend, elucidate, and convey mathematical notions with utmost precision.", +"postfix_prompt": "Infer the constants from asymptote code (if has) first and then solve the question." +``` +### \ No newline at end of file From 8b73a12af337f732869034dce7d5587db905a3ac Mon Sep 17 00:00:00 2001 From: Sayo Date: Sat, 21 Oct 2023 19:06:56 +0800 Subject: [PATCH 08/56] [feat] update expertise_prompt_agent prompt --- creator/agents/expertise_prompt_agent.py | 24 ++++--------------- .../prompts/expertise_prompt_agent_prompt.md | 19 ++++----------- 2 files changed, 9 insertions(+), 34 deletions(-) diff --git a/creator/agents/expertise_prompt_agent.py b/creator/agents/expertise_prompt_agent.py index a4e8904..a44d4d8 100644 --- a/creator/agents/expertise_prompt_agent.py +++ b/creator/agents/expertise_prompt_agent.py @@ -7,7 +7,7 @@ from creator.utils import load_system_prompt from creator.llm.llm_creator import create_llm -from base import BaseAgent +from creator.agents.base import BaseAgent class ExpertisePromptAgent(BaseAgent): @@ -40,16 +40,16 @@ def create_expertise_prompt_agent(llm): function_schema = { "name": "expertise_prompt", - "description": "a function that guide GPT to act as an expert in the field to respond to the user's request", + "description": "a function that guide GPT to act as a specific expert in the field to respond to the user's request", "parameters": { "properties": { "prefix_prompt": { "type": "string", - "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset.", + "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset within a particular vertical.", }, "postfix_prompt": { "type": "string", - "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response.", + "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response. If you don't know how to give guidance based on the user's request just let the GPT think step-by-step.", } }, "type": "object", @@ -67,18 +67,4 @@ def create_expertise_prompt_agent(llm): llm = create_llm(config) -expertise_prompt_agent = create_expertise_prompt_agent(llm) - -if __name__ == "__main__": - TabMWP_test = """\"question\": \"Hannah baked cookies each day for a bake sale. How many more cookies did Hannah bake on Saturday than on Sunday?\", \"table\": \"Name: Cookies baked\nUnit: cookies\nContent:\nDay | Number of cookies\nFriday | 163\nSaturday | 281\nSunday | 263\" -""" - tool_transfer_test=""""intro": "The tool is used to calculate the optimal number of units to produce to maximize profit for a manufacturing company. It takes into account the fixed costs, variable costs, selling price, and demand for the product. The function uses the formula Profit = (Selling Price * Quantity) - (Variable Cost * Quantity) - Fixed Cost to calculate the profit and returns the optimal quantity to produce.", "tool": "```python\ndef calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand):\n \"\"\"\n Calculates the optimal number of units to produce to maximize profit.\n\n Parameters:\n selling_price (float): The price at which the product is sold.\n variable_cost (float): The cost of producing one unit of the product.\n fixed_cost (float): The fixed cost of production.\n demand (int): The number of units that can be sold at the given price.\n\n Returns:\n int: The optimal number of units to produce to maximize profit.\n \"\"\"\n # Calculate the profit for each quantity\n profits = []\n for quantity in range(1, demand+1):\n profit = (selling_price * quantity) - (variable_cost * quantity) - fixed_cost\n profits.append(profit)\n\n # Find the quantity that maximizes profit\n optimal_quantity = profits.index(max(profits)) + 1\n\n # Return the optimal quantity\n return optimal_quantity\n```", "scn1": "Production Planning\nA manufacturing company produces a product that has a fixed cost of $10,000, a variable cost of $5 per unit, and a selling price of $20 per unit. The company can sell up to 5,000 units of the product at this price. What is the optimal number of units to produce to maximize profit?", "sol1": "```python\n# Set the inputs for the manufacturing company\nselling_price = 20\nvariable_cost = 5\nfixed_cost = 10000\ndemand = 5000\n\n# Calculate the optimal number of units to produce\noptimal_quantity = calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand)\n\n# Print the optimal quantity\nprint(optimal_quantity)\n```", "ans1": 5000.0, "scn2": "Pricing Strategy\nA company produces a product that has a fixed cost of $20,000, a variable cost of $10 per unit, and a demand of 10,000 units. The company wants to maximize profit and is considering two pricing strategies. The first strategy is to sell the product at $30 per unit, and the second strategy is to sell the product at $35 per unit. What is the optimal pricing strategy for the company?", "sol2": "```python\n# Set the inputs for the company\nfixed_cost = 20000\nvariable_cost = 10\ndemand = 10000\n\n# Strategy 1: Selling the product at $30 per unit\nselling_price_1 = 30\noptimal_quantity_1 = calculate_optimal_units(selling_price_1, variable_cost, fixed_cost, demand)\nprofit_1 = (selling_price_1 * optimal_quantity_1) - (variable_cost * optimal_quantity_1) - fixed_cost\n\n# Strategy 2: Selling the product at $35 per unit\nselling_price_2 = 35\noptimal_quantity_2 = calculate_optimal_units(selling_price_2, variable_cost, fixed_cost, demand)\nprofit_2 = (selling_price_2 * optimal_quantity_2) - (variable_cost * optimal_quantity_2) - fixed_cost\n\n# Determine the optimal pricing strategy\nif profit_1 > profit_2:\n print(\"The optimal pricing strategy is to sell the product at $30 per unit.\")\nelse:\n print(\"The optimal pricing strategy is to sell the product at $35 per unit.\")\n```", "ans2": 35.0, "scn3": "Capacity Planning\nA company produces a product that has a fixed cost of $50,000, a variable cost of $15 per unit, and a selling price of $25 per unit. The company has a production capacity of 10,000 units. What is the optimal number of units to produce to maximize profit?", "sol3": "```python\n# Set the inputs for the company\nselling_price = 25\nvariable_cost = 15\nfixed_cost = 50000\ndemand = 10000\n\n# Calculate the optimal number of units to produce\noptimal_quantity = calculate_optimal_units(selling_price, variable_cost, fixed_cost, demand)\n\n# Print the optimal quantity\nprint(optimal_quantity)\n```" -""" - messages = [{ - "role": "user", - "content": TabMWP_test - }] - result = expertise_prompt_agent.run({ - "messages": messages, - "verbose": True, - }) +expertise_prompt_agent = create_expertise_prompt_agent(llm) \ No newline at end of file diff --git a/creator/prompts/expertise_prompt_agent_prompt.md b/creator/prompts/expertise_prompt_agent_prompt.md index 75fa62f..41e0a9e 100644 --- a/creator/prompts/expertise_prompt_agent_prompt.md +++ b/creator/prompts/expertise_prompt_agent_prompt.md @@ -1,15 +1,4 @@ -You are a "Prompt Advisor". Your task is to guide GPT to respond as an expert in a given field based on the user's request. Provide a short prefix_prompt as specific instructions for the expert role. Then, offer brief tips in a postfix_prompt. The user's original request will be placed between the two prompts. - -Here is an example: -### -User_Request: -``` -A sphere is inscribed in a cone with height 4 and base radius 3. What is the ratio of the volume of the sphere to the volume of the cone?\n\n[asy]\nfill(circle((1.5,0),1.5),gray(.7));\ndraw((0,3)--(4,0)--(0,-3));\ndraw(circle((1.5,0),1.5));\ndraw((0,0)..(1.5,-.3)..(3,0));\ndraw((0,0)..(1.5,.3)..(3,0),dashed);\ndraw(xscale(.15)*circle((0,0),3));\n[/asy] -``` - -result: -``` -"prefix_prompt": "You are doing a math test as a mathematically-trained expert, well-versed in a wide range of mathematical concepts and terms; your main objective is to comprehend, elucidate, and convey mathematical notions with utmost precision.", -"postfix_prompt": "Infer the constants from asymptote code (if has) first and then solve the question." -``` -### \ No newline at end of file +You are a "Prompt Advisor". Your task is to guide GPT to respond as an expert in a specific vertical within a given field based on the user's request. Pay attention to analyzing real user requests and give reasonable guidance. +For the prefix_prompt, provide a specific concise directive that instructs GPT to playing a particular expert role or mindset. For instance, If the user's request is related to python code, you can have GPT play the role of "You are a skilled python programmer" instead of "You are a programmer". +Then, offer brief tips or guidelines in the postfix_prompt to give additional context or direction for GPT's response. +The user's original request will be placed between the two prompts. \ No newline at end of file From 4c31d0abb400bfae8de895d8bdf4e65b665635ab Mon Sep 17 00:00:00 2001 From: Gong Junmin <30956809+ChuxiJ@users.noreply.github.com> Date: Sun, 22 Oct 2023 16:12:39 +0800 Subject: [PATCH 09/56] Create main.yml --- .github/workflows/main.yml | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) create mode 100644 .github/workflows/main.yml diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml new file mode 100644 index 0000000..2b4b5b2 --- /dev/null +++ b/.github/workflows/main.yml @@ -0,0 +1,25 @@ +name: ci +on: + push: + branches: + - master + - main +permissions: + contents: write +jobs: + deploy: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@v4 + - uses: actions/setup-python@v4 + with: + python-version: 3.x + - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV + - uses: actions/cache@v3 + with: + key: mkdocs-material-${{ env.cache_id }} + path: .cache + restore-keys: | + mkdocs-material- + - run: pip install mkdocs-material + - run: mkdocs gh-deploy --force From eeb1b6cc5b4601f64364d1cafaef67740e323bcb Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Sun, 22 Oct 2023 16:13:42 +0800 Subject: [PATCH 10/56] feat: add material theme --- docs/api_doc.md | 46 +++++++++++++++---------------- docs/pics/logo.png | Bin 0 -> 2432 bytes mkdocs.yml | 67 +++++++++++++++++++++++++++++++++++++++++++-- 3 files changed, 88 insertions(+), 25 deletions(-) create mode 100644 docs/pics/logo.png diff --git a/docs/api_doc.md b/docs/api_doc.md index eca7baa..628a610 100644 --- a/docs/api_doc.md +++ b/docs/api_doc.md @@ -1,7 +1,7 @@ ## Open-Creator API Documentation -### Function: `create` -Generates a `CodeSkill` instance using different input sources. +### Function: `#!python create` +Generates a `#!python CodeSkill` instance using different input sources. #### Parameters: - `request`: String detailing the skill functionality. @@ -12,46 +12,46 @@ Generates a `CodeSkill` instance using different input sources. - `huggingface_skill_path`: Path to the skill within the Huggingface repository. #### Returns: -- `CodeSkill`: The created skill. +- `#!python CodeSkill`: The created skill. #### Usage: 1. Creating Skill using a Request String: -```python +``` py skill = create(request="filter how many prime numbers are in 201") ``` 2. Creating Skill using Messages: - Directly: -```python +``` py skill = create(messages=[{"role": "user", "content": "write a program..."}]) ``` - Via JSON Path: -```python +``` py skill = create(messages_json_path="./messages_example.json") ``` 3. Creating Skill using File Content or File Path: - Direct Content: -```python +``` py skill = create(file_content="def example_function(): pass") ``` - File Path: -```python +``` py skill = create(file_path="../creator/utils/example.py") ``` 4. Creating Skill using Skill Path or Skill JSON Path: - JSON Path: -```python +``` py skill = create(skill_json_path="~/.cache/open_creator/skill_library/create/skill.json") ``` - Skill Path: -```python +``` py skill = create(skill_path="~/.cache/open_creator/skill_library/create") ``` 5. Creating Skill using Huggingface Repository ID and Skill Path: If a skill is hosted in a Huggingface repository, you can create it by specifying the repository ID and the skill path within the repository. -```python +``` py skill = create(huggingface_repo_id="YourRepo/skill-library", huggingface_skill_path="specific_skill") ``` @@ -64,7 +64,7 @@ skill = create(huggingface_repo_id="YourRepo/skill-library", huggingface_skill_p ### Function: `save` -Stores a `CodeSkill` instance either to a local path or a Huggingface repository. In default just use `save(skill)` and it will store the skill into the default path. Only save the skill when the user asks to do so. +Stores a `#!python CodeSkill` instance either to a local path or a Huggingface repository. In default just use `save(skill)` and it will store the skill into the default path. Only save the skill when the user asks to do so. #### Parameters: - `skill` (CodeSkill): The skill instance to be saved. @@ -75,15 +75,15 @@ Stores a `CodeSkill` instance either to a local path or a Huggingface repository - None #### Usage: -The `save` function allows for the persistent storage of a `CodeSkill` instance by saving it either locally or to a specified Huggingface repository. +The `save` function allows for the persistent storage of a `#!python CodeSkill` instance by saving it either locally or to a specified Huggingface repository. 1. **Save to Huggingface Repository:** -```python +``` py save(skill=skill, huggingface_repo_id="YourRepo/skill_library") ``` 2. **Save Locally:** -```python +``` py save(skill=skill, skill_path="/path/to/save") ``` @@ -101,18 +101,18 @@ Retrieve skills related to a specified query from the available pool of skills. - `threshold` (Optional[float]): Minimum similarity score to return a skill. Default is 0.8. #### Returns: -- List[CodeSkill]: A list of retrieved `CodeSkill` objects that match the query. +- List[CodeSkill]: A list of retrieved `#!python CodeSkill` objects that match the query. #### Usage: The `search` function allows users to locate skills related to a particular query string. This is particularly useful for identifying pre-existing skills within a skill library that may fulfill a requirement or for exploring available functionalities. 1. **Basic Search:** -```python +``` py skills = search("extract pages from a pdf") ``` 2. **Refined Search:** -```python +``` py skills = search("extract pages from a pdf", top_k=3, threshold=0.85) ``` @@ -131,7 +131,7 @@ Execute a skill with provided arguments or request. - **Example Usage**: -```python +``` py linenums="1" skills = search("pdf extract section") if skills: skill = skills[0] @@ -149,7 +149,7 @@ Validate a skill using a tester agent. - **Example Usage**: -```python +``` py linenums="1" skill = create(request="filter prime numbers in a range, e.g., filter_prime_numbers(2, 201)") test_summary = skill.test() print(test_summary) @@ -161,17 +161,17 @@ Modify and refine skills using operator overloading. 1. **Combining Skills**: Utilize the `+` operator to chain or execute skills in parallel, detailing the coordination with the `>` operator. -```python +``` py new_skill = skillA + skillB > "Explanation of how skills A and B operate together" ``` 2. **Refactoring Skills**: Employ the `>` operator to enhance or modify existing skills. - ```python + ``` py refactored_skill = skill > "Descriptive alterations or enhancements" ``` 3. **Decomposing Skills**: Use the `<` operator to break down a skill into simpler components. - ```python + ``` py simpler_skills = skill < "Description of how the skill should be decomposed" ``` diff --git a/docs/pics/logo.png b/docs/pics/logo.png new file mode 100644 index 0000000000000000000000000000000000000000..f1ab997bf31ead37b33e521908254c52beb34e78 GIT binary patch literal 2432 zcmb_edpuNm8$UCSW@zOwMI|~jhL)JYD!0Lm#xSmtw~!Rg$aO|D6JxV3({6aR8WWYN zux^D`n%U@LbzRpebosbdh_Vu)5Su+yz4gAI_s{n|=W~9~?fZPc&-451WNh_vQBz&3 z3IKo_)0OUv*e1n?U51>$`d8&62F3SvaRRE^*Nq~zWKRee2x4u?NLEzH<|H%n z7jC4LqR85M^CH-YvVz~iobnO*AGyyu6tY76--h|L>BlUxs)H(p{N>slRM!L#U;qG< zz@*dsgsAbJ-nd@dZ=5@37MC1nt#u>yH`X&iSFzlC>Ss{<-CFUPq{O*20F8xH{^#RH zCzxdV!~5sAm+BYot_@#ky)KWMTO0##B%OKaJ`s(->8{f+#3gX3HuEcKUg%dwcjTQa z(?{-4KUSGK-SNUq>CwS86$S7CQyrf6(l4|MWIzCEFQEhUb(a) z{xaN(YI=>Q#hE#fcW4g9&Zrly%ScRwSbJr%)s6npaAwn#2pZb>P21Kxm#6iS-@p0_?VnS>d#%>{Snw7LJgbs}Zi<;qrLB z00|yZh@&q|SXB_0-ntZ{!3G1fKG590LCsyj*7wT(MZpC&?oPI39!3;~(yS_PWfR}% z5-Z2v88P>ytAu`^bJr`pqu3++HKJv?uT*95JKGlge~oQtxn(FGFlst43n%-Vict3T?WC*fVaoaqjg5^tz$?~Ln_ox6`3g^G@x z!sMz^;RB+RdP;&ip-T6*JD`!?>fuIB*HdtrOSJxuKzlyYyV0OE;Hjb5g646wc41J- z8_V+4z&&T`*VS!KokV6luUp;Jwa-HXN0c3UMDk97hj+dNvOy>A#lB$Yn)aRou!t-7 zfNZurL!m^JlpzD~76&P*?9fVW>NUX;D>IR8Q(%MMvX>@ku$d5pOz%{GXzqiL#q!e3 ztoM4MqMWRs28&f7qO|(0I}UaoeP`q#rA|-4^~vxY8i#VS>&{w^q78i8X;gREUal!(6d(ugI9 zweZ}iOB4Y#!nr{HS~Qqfmb*femI7<|{c59z5_xpFVKQK<7+dk^^kHh3V@0&z;>B30 zFvreynsN#rwkUk&f!sn#%mp5Z^{ry7wiBshkDhnNF380OmVGTsmrQCjCbw~~e49hI zCbk;7&IegjiO3Y4m*Nu8pwUw&Oga(bJ*`>w>Yyv@#rfYiK*#suSVy}zp)qLC26zLd z+M`6TA}%wYQQ=olVtN~A#Y(a2r*PP|YL(8q-CCzJ09x3eWpc|6kZpcVWPjKu?kQy} z^uYWk8cQt|17fFutP>D+y^l_4ZGh~Nl9SiB?MTrFC z@-~+Z;;_MGw^pvP*BE+yLK=W$EmF@_>!v6(t4{0FXh(H}j^;53>$ zhc4I-{kcmPDZ0DbU|%GoFf(`O+cH=AX0Pj8Yfnqzu6?bjd(g|Uk?p@3@yZTHCDueS zu5r91n>3RKbNZ)8wZ`bPiGn0_zgj2J$i)?$b8b)J~Of2@C4^#**&RJ zKHls|c`v>%^33SSx&mF}V_Q~jk=P(4D{l|)zbta)5rQu{+Oo8Mz?AD6Kg8fxclM0m ztv?D{B^M5-wGCANl+7xfYdCHpSe8c0B3^7>x99D^v{hm%aShyHZjjRyoIzk}9=o4a zdAxa7o)fdvygjCtlzYYZ4@pUkuig3WB<`@PAv4Xtwp+$}9)0 Date: Mon, 23 Oct 2023 10:41:26 +0800 Subject: [PATCH 11/56] feat: mkdocs add primary black and add social footer --- mkdocs.yml | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/mkdocs.yml b/mkdocs.yml index 7bce553..1b40141 100644 --- a/mkdocs.yml +++ b/mkdocs.yml @@ -26,7 +26,6 @@ theme: icon: repo: fontawesome/brands/github palette: - # Palette toggle for automatic mode - media: "(prefers-color-scheme)" toggle: @@ -43,6 +42,7 @@ theme: # Palette toggle for dark mode - media: "(prefers-color-scheme: dark)" scheme: slate + primary: black toggle: icon: material/brightness-4 name: Switch to system preference @@ -50,7 +50,6 @@ theme: font: text: Roboto code: Roboto Mono - features: - content.code.copy - content.code.select @@ -70,6 +69,7 @@ theme: - search.suggest - search.highlight - search.share + - navigation.footer plugins: - search @@ -89,3 +89,12 @@ markdown_extensions: - pymdownx.inlinehilite - pymdownx.snippets - pymdownx.superfences + +extra: + social: + - icon: fontawesome/brands/twitter + link: https://twitter.com/UseOpenCreator + - icon: fontawesome/brands/github + link: https://github.com/timedomain-tech/open-creator + - icon: fontawesome/brands/discord + link: https://discord.gg/eEraZEry53 From 49c0e0f1005cefd04bd0a8206efc5986402d83ca Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Mon, 23 Oct 2023 11:51:31 +0800 Subject: [PATCH 12/56] feat: test prompt enhancer agent --- creator/agents/expertise_prompt_agent.py | 70 ------------------- creator/agents/prompt_enhancer_agent.py | 56 +++++++++++++++ creator/config/library.py | 3 +- creator/core/core.py | 14 +++- ...mpt.md => prompt_enhancer_agent_prompt.md} | 0 creator/prompts/prompt_enhancer_schema.json | 18 +++++ creator/utils/printer.py | 5 +- docs/index.md | 3 - 8 files changed, 93 insertions(+), 76 deletions(-) delete mode 100644 creator/agents/expertise_prompt_agent.py create mode 100644 creator/agents/prompt_enhancer_agent.py rename creator/prompts/{expertise_prompt_agent_prompt.md => prompt_enhancer_agent_prompt.md} (100%) create mode 100644 creator/prompts/prompt_enhancer_schema.json diff --git a/creator/agents/expertise_prompt_agent.py b/creator/agents/expertise_prompt_agent.py deleted file mode 100644 index a44d4d8..0000000 --- a/creator/agents/expertise_prompt_agent.py +++ /dev/null @@ -1,70 +0,0 @@ -from typing import Any, Dict - -from langchain.prompts import ChatPromptTemplate -from langchain.output_parsers.json import parse_partial_json - -from creator.config.library import config -from creator.utils import load_system_prompt -from creator.llm.llm_creator import create_llm - -from creator.agents.base import BaseAgent - - -class ExpertisePromptAgent(BaseAgent): - output_key: str = "expertise_prompt" - - @property - def _chain_type(self): - return "ExpertisePromptAgent" - - def construct_prompt(self, langchain_messages: Dict[str, Any]): - prompt = ChatPromptTemplate.from_messages(messages=[ - *langchain_messages, - ("system", self.system_template) - ]) - return prompt - - def parse_output(self, messages): - function_call = messages[-1].get("function_call", None) - if function_call is not None: - rewrited_prompt = parse_partial_json(function_call.get("arguments", "{}")) - prefix_prompt = rewrited_prompt.get("prefix_prompt", "") - postfix_prompt = rewrited_prompt.get("postfix_prompt", "") - result = {"prefix_prompt": prefix_prompt, "postfix_prompt": postfix_prompt} - return {"expertise_prompt": result} - return {"expertise_prompt": None} - - -def create_expertise_prompt_agent(llm): - template = load_system_prompt(config.expertise_prompt_agent_prompt_path) - - function_schema = { - "name": "expertise_prompt", - "description": "a function that guide GPT to act as a specific expert in the field to respond to the user's request", - "parameters": { - "properties": { - "prefix_prompt": { - "type": "string", - "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset within a particular vertical.", - }, - "postfix_prompt": { - "type": "string", - "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response. If you don't know how to give guidance based on the user's request just let the GPT think step-by-step.", - } - }, - "type": "object", - "required": ["prefix_prompt", "postfix_prompt"] - } - } - - chain = ExpertisePromptAgent( - llm=llm, - system_template=template, - function_schemas=[function_schema], - verbose=False - ) - return chain - - -llm = create_llm(config) -expertise_prompt_agent = create_expertise_prompt_agent(llm) \ No newline at end of file diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py new file mode 100644 index 0000000..aa139ed --- /dev/null +++ b/creator/agents/prompt_enhancer_agent.py @@ -0,0 +1,56 @@ +from typing import Any, Dict + +from langchain.prompts import ChatPromptTemplate +from langchain.output_parsers.json import parse_partial_json + +from creator.config.library import config +from creator.utils import load_system_prompt +from creator.llm.llm_creator import create_llm +from creator.utils import print + +from creator.agents.base import BaseAgent +import json + + +class PromptEnhancerAgent(BaseAgent): + output_key: str = "enhanced_prompt" + + @property + def _chain_type(self): + return "PromptEnhancerAgent" + + def construct_prompt(self, langchain_messages: Dict[str, Any]): + prompt = ChatPromptTemplate.from_messages(messages=[ + *langchain_messages, + ("system", self.system_template) + ]) + return prompt + + def parse_output(self, messages): + orignal_request = messages[0].get("content", "") + function_call = messages[1].get("function_call", None) + if function_call is not None: + rewrited_prompt = parse_partial_json(function_call.get("arguments", "{}")) + prefix_prompt = rewrited_prompt.get("prefix_prompt", "") + postfix_prompt = rewrited_prompt.get("postfix_prompt", "") + print(f"[green]{prefix_prompt}[/green]\n{orignal_request}\n[green]{postfix_prompt}[/green]") + return {"enhanced_prompt": "\n".join([prefix_prompt, orignal_request, postfix_prompt])} + return {"enhanced_prompt": orignal_request} + + +def create_prompt_enhancer_agent(llm): + template = load_system_prompt(config.prompt_enhancer_agent_prompt_path) + + with open(config.prompt_enhancer_schema_path, encoding="utf-8") as f: + function_schema = json.load(f) + + chain = PromptEnhancerAgent( + llm=llm, + system_template=template, + function_schemas=[function_schema], + verbose=False + ) + return chain + + +llm = create_llm(config) diff --git a/creator/config/library.py b/creator/config/library.py index 9b53611..7bf2897 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -86,7 +86,8 @@ class LibraryConfig(BaseModel): tips_for_debugging_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_debugging_prompt.md") tips_for_testing_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_testing_prompt.md") tips_for_veryfy_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_veryfy_prompt.md") - expertise_prompt_agent_prompt_path: str = os.path.join(project_dir, "prompts", "expertise_prompt_agent_prompt.md") + prompt_enhancer_agent_prompt_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_agent_prompt.md") + prompt_enhancer_schema_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_schema.json") use_rich: bool = True use_file_logger: bool = False diff --git a/creator/core/core.py b/creator/core/core.py index 9f09bd0..90b55c7 100644 --- a/creator/core/core.py +++ b/creator/core/core.py @@ -1,6 +1,8 @@ import os from typing import Union, List, Optional from creator.agents import skill_extractor_agent, code_interpreter_agent +from creator.agents.prompt_enhancer_agent import create_prompt_enhancer_agent +from creator.llm.llm_creator import create_llm from creator.core.skill import CodeSkill, BaseSkill, BaseSkillMetadata from creator.config.library import config from creator.utils import print @@ -144,11 +146,21 @@ def create( return cls._create_from_skill_json_path(skill_json_path) if request: - messages = code_interpreter_agent.run({ + config.use_rich = False + prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(config)) + enhanced_prompt = prompt_enhancer_agent.run({ "messages": [{ "role": "user", "content": request }], + "verbose": False, + }) + config.use_rich = True + messages = code_interpreter_agent.run({ + "messages": [{ + "role": "user", + "content": enhanced_prompt + }], "verbose": True, }) diff --git a/creator/prompts/expertise_prompt_agent_prompt.md b/creator/prompts/prompt_enhancer_agent_prompt.md similarity index 100% rename from creator/prompts/expertise_prompt_agent_prompt.md rename to creator/prompts/prompt_enhancer_agent_prompt.md diff --git a/creator/prompts/prompt_enhancer_schema.json b/creator/prompts/prompt_enhancer_schema.json new file mode 100644 index 0000000..8ec27ec --- /dev/null +++ b/creator/prompts/prompt_enhancer_schema.json @@ -0,0 +1,18 @@ +{ + "name": "prompt_enhancer", + "description": "a function that guide GPT to act as a specific expert in the field to respond to the user's request", + "parameters": { + "properties": { + "prefix_prompt": { + "type": "string", + "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset within a particular vertical." + }, + "postfix_prompt": { + "type": "string", + "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response. If you don't know how to give guidance based on the user's request just let the GPT think step-by-step." + } + }, + "type": "object", + "required": ["prefix_prompt", "postfix_prompt"] + } +} \ No newline at end of file diff --git a/creator/utils/printer.py b/creator/utils/printer.py index 6e62d99..e325416 100644 --- a/creator/utils/printer.py +++ b/creator/utils/printer.py @@ -51,7 +51,10 @@ def add_default_callback(self): def default_print(message, end='\n', file=None, flush=False, output_option='terminal'): target_file = file or self.output_capture if output_option in ['terminal', 'both']: - console = Console(force_jupyter=self.is_jupyter, force_terminal=self.is_terminal, force_interactive=self.is_interactive, file=target_file) + if self.is_jupyter: + console = Console(force_jupyter=self.is_jupyter, force_terminal=self.is_terminal, force_interactive=self.is_interactive, file=target_file) + else: + console = Console(force_jupyter=self.is_jupyter, force_terminal=self.is_terminal, force_interactive=self.is_interactive) console.print(message, end=end) # if output_option in ['stdout', 'both']: # rich_print(message, end=end, file=sys.stdout, flush=flush) diff --git a/docs/index.md b/docs/index.md index 53e5c88..79f9c45 100644 --- a/docs/index.md +++ b/docs/index.md @@ -33,6 +33,3 @@ ## Framework ![](tech_report/figures/framework.png) - ---- - From e9553e9bd63d16dc0fda0a1c162cfc7970f1b23a Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Mon, 23 Oct 2023 14:16:56 +0800 Subject: [PATCH 13/56] feat: update prompt for prompt_enhancer_agent --- creator/prompts/prompt_enhancer_agent_prompt.md | 9 +++++---- creator/prompts/prompt_enhancer_schema.json | 6 +++--- 2 files changed, 8 insertions(+), 7 deletions(-) diff --git a/creator/prompts/prompt_enhancer_agent_prompt.md b/creator/prompts/prompt_enhancer_agent_prompt.md index 41e0a9e..979762d 100644 --- a/creator/prompts/prompt_enhancer_agent_prompt.md +++ b/creator/prompts/prompt_enhancer_agent_prompt.md @@ -1,4 +1,5 @@ -You are a "Prompt Advisor". Your task is to guide GPT to respond as an expert in a specific vertical within a given field based on the user's request. Pay attention to analyzing real user requests and give reasonable guidance. -For the prefix_prompt, provide a specific concise directive that instructs GPT to playing a particular expert role or mindset. For instance, If the user's request is related to python code, you can have GPT play the role of "You are a skilled python programmer" instead of "You are a programmer". -Then, offer brief tips or guidelines in the postfix_prompt to give additional context or direction for GPT's response. -The user's original request will be placed between the two prompts. \ No newline at end of file +You are a "Prompt Advisor", guiding Large Language Models (LLMs) to emulate expertise in specific niches. Read user request carefully, only use `prompt_enhancer` tool to reply the user. + +## Tools +### prompt_enhancer +When you send a message to prompt_enhancer, it will use `"\n".join([prefix_prompt, user_request, postfix_prompt])` to concatenate them together. The user's original request will be placed between the two prompts. Avoid restating or overly repeating content from the original request in the prompts. Ensure the user's intent remains intact between prompts. If nothing to add, leave the prefix and postfix as blank strings. \ No newline at end of file diff --git a/creator/prompts/prompt_enhancer_schema.json b/creator/prompts/prompt_enhancer_schema.json index 8ec27ec..4e6e4cd 100644 --- a/creator/prompts/prompt_enhancer_schema.json +++ b/creator/prompts/prompt_enhancer_schema.json @@ -1,15 +1,15 @@ { "name": "prompt_enhancer", - "description": "a function that guide GPT to act as a specific expert in the field to respond to the user's request", + "description": "Function guiding LLMs to act as a niche expert in response to user queries.", "parameters": { "properties": { "prefix_prompt": { "type": "string", - "description": "A concise directive provided at the beginning of the user's request, guiding GPT to adopt a specific expert role or mindset within a particular vertical." + "description": "Initial directive setting LLM's expert role. e.g., 'You are a skilled python programmer' over 'You are a programmer'." }, "postfix_prompt": { "type": "string", - "description": "A brief set of tips or guidelines placed after the user's original request, offering additional context or direction for GPT's response. If you don't know how to give guidance based on the user's request just let the GPT think step-by-step." + "description": "Tips or context following the user's query. If unsure about guidance, let LLMs think sequentially." } }, "type": "object", From 232e424ab80fe85ffe2c068090b142eaea94d3da Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 12:06:39 +0800 Subject: [PATCH 14/56] fix: repl --- creator/app/streamlit_app.py | 4 +- creator/client/command.py | 4 +- creator/client/repl/__init__.py | 5 +- creator/client/repl/app.py | 138 +++++++---------- creator/client/repl/completer.py | 1 + creator/client/repl/constants.py | 20 ++- creator/client/repl/handler.py | 144 +++++++++--------- creator/client/repl/lexer.py | 38 ----- creator/client/repl/style.py | 4 - .../prompts/prompt_enhancer_agent_prompt.md | 4 +- pyproject.toml | 4 +- 11 files changed, 152 insertions(+), 214 deletions(-) delete mode 100644 creator/client/repl/lexer.py diff --git a/creator/app/streamlit_app.py b/creator/app/streamlit_app.py index 43a3c33..0d81bfe 100644 --- a/creator/app/streamlit_app.py +++ b/creator/app/streamlit_app.py @@ -28,7 +28,9 @@ def setup_slidebar(): openai_api_key = st.text_input("OpenAI API Key", key="chatbot_api_key", type="password") "[Get an OpenAI API key](https://platform.openai.com/account/api-keys)" "[View the source code](https://github.com/timedomain-tech/open-creator/tree/main/creator/app/streamlit_app.py)" - os.environ["OPENAI_API_KEY"] = openai_api_key + curr_api_key = os.environ.get("OPENAI_API_KEY", "") + if not curr_api_key: + os.environ["OPENAI_API_KEY"] = openai_api_key model_list = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"] model = st.selectbox("Model", model_list, key="model") config.model = model diff --git a/creator/client/command.py b/creator/client/command.py index f82769a..48c24aa 100644 --- a/creator/client/command.py +++ b/creator/client/command.py @@ -1,6 +1,7 @@ import argparse from rich.rule import Rule import json +import asyncio from creator.utils.printer import print as rich_print from creator.config.open_config import open_user_config @@ -216,7 +217,8 @@ def cmd_client(): return if not args.command or args.interactive: - repl_app.run(args.quiet) + loop = asyncio.get_event_loop() + loop.run_until_complete(repl_app.run(args.quiet)) return if args.command == "create": diff --git a/creator/client/repl/__init__.py b/creator/client/repl/__init__.py index 984f1ad..6864526 100644 --- a/creator/client/repl/__init__.py +++ b/creator/client/repl/__init__.py @@ -1,9 +1,6 @@ from .app import OpenCreatorREPL -from .handler import RequestHandler -handler = RequestHandler() - -repl_app = OpenCreatorREPL(handler) +repl_app = OpenCreatorREPL() __all__ = ["repl_app"] diff --git a/creator/client/repl/app.py b/creator/client/repl/app.py index beed19c..91db315 100644 --- a/creator/client/repl/app.py +++ b/creator/client/repl/app.py @@ -1,108 +1,82 @@ -from prompt_toolkit.application import Application from prompt_toolkit.key_binding import KeyBindings from prompt_toolkit.keys import Keys -from prompt_toolkit.layout import HSplit, Layout -from prompt_toolkit.widgets import TextArea -from prompt_toolkit.layout.dimension import Dimension from prompt_toolkit.auto_suggest import AutoSuggestFromHistory -from prompt_toolkit.lexers import PygmentsLexer +from prompt_toolkit.shortcuts import PromptSession +from prompt_toolkit.document import Document from pygments.lexers.markup import MarkdownLexer -from prompt_toolkit.filters import to_filter - +from prompt_toolkit.lexers import PygmentsLexer import traceback -from .constants import prompt_message, help_text +from .constants import help_text, prompt_message, interpreter_message from .completer import completer, file_history from .style import style -from .lexer import CustomLexer +from .handler import RequestHandler +from questionary import Question +import sys class OpenCreatorREPL: + interpreter = False - def __init__(self, accept_callback=None): - self.accept_callback = accept_callback - - def run(self, quiet=False): - output_text = "" if quiet else help_text - self.output_field = TextArea(text=output_text, height=Dimension(min=0, weight=1), focusable=True, read_only=True, focus_on_click=True, lexer=CustomLexer(), scrollbar=True) - self.input_field = TextArea( - height=Dimension(min=1, weight=100), - prompt=prompt_message, - multiline=True, - wrap_lines=False, - focus_on_click=True, - dont_extend_height=False, - completer=completer, - auto_suggest=AutoSuggestFromHistory(), - history=file_history, - lexer=PygmentsLexer(MarkdownLexer), - complete_while_typing=True, - ) - self.input_field.buffer.enable_history_search = to_filter(True) - self.input_field.accept_handler = self.accept - + def setup(self): + self.handler = RequestHandler() # The key bindings. kb = KeyBindings() - @kb.add(Keys.ControlD) - @kb.add(Keys.ControlQ) - def _(event): - " Pressing Ctrl-Q will exit the user interface. " - self.accept_callback.handle_exit(event.app, self.output_field.text) - # emacs control + j new line keybindings @kb.add(Keys.ControlJ) def _(event): event.current_buffer.insert_text('\n') - @kb.add(Keys.ControlC) - def _(event): - buffer = event.app.current_buffer - self.input_field.accept_handler(buffer, keyboard_interrupt=True) - event.app.current_buffer.reset() - - @kb.add(Keys.Enter) - def _(event): - " When enter is pressed, we insert a newline. " - buffer = event.app.current_buffer - if self.input_field.text.startswith("%exit"): - self.accept_callback.handle_exit(event.app, self.output_field.text) - return - - self.input_field.accept_handler(buffer) - event.app.current_buffer.reset() - - container = HSplit( - [ - self.output_field, - self.input_field, - ] - ) - # Run application. - self.application = Application( - layout=Layout(container, focused_element=self.input_field), - key_bindings=kb, + self.prompt_session = PromptSession( + prompt_message if not self.interpreter else interpreter_message, style=style, + multiline=False, + lexer=PygmentsLexer(MarkdownLexer), + history=file_history, + completer=completer, + auto_suggest=AutoSuggestFromHistory(), + complete_while_typing=True, mouse_support=True, - full_screen=True, + key_bindings=kb ) - self.application.run() - - def accept(self, buff, keyboard_interrupt=False): - - self.output_field.buffer.read_only = to_filter(False) - show_stderr = True - if keyboard_interrupt: - output = "KeyboardInterrupt" - else: + async def ask(self, interpreter=False): + if self.interpreter != interpreter: + self.interpreter = interpreter + self.setup() + self.prompt_session.default_buffer.reset(Document()) + question = Question(self.prompt_session.app) + output = await question.unsafe_ask_async(patch_stdout=True) + return output + + async def run(self, quiet=False, interpreter=False): + self.setup() + if not quiet: + await self.handler.show_output("", help_text, add_prompt_prefix=False) + + while 1: try: - self.accept_callback.handle(self.input_field.text, self.output_field) - show_stderr = False + user_request = await self.ask(interpreter) + user_request = user_request.strip() + await self.handler.show_output(user_request, "", add_newline=user_request != "", interpreter=interpreter) + if user_request == "%exit": + sys.exit() + if user_request.startswith("%interpreter"): + interpreter = not interpreter + mode = "on" if interpreter else "off" + await self.handler.show_output("", f"[red]Toggled Interpreter mode {mode}![/red]", add_prompt_prefix=False, add_newline=False, add_request=False) + continue + if user_request: + await self.handler.handle(user_request, interpreter=interpreter) + except KeyboardInterrupt: + user_request = self.prompt_session.default_buffer.text + await self.handler.show_output(user_request, "", grey=True, interpreter=interpreter) + await self.handler.show_output("", "[red]KeyboardInterrupt[/red]", add_prompt_prefix=False, add_newline=False, add_request=False) + except EOFError: + sys.exit() except Exception: - output = f"{traceback.format_exc()}" - if self.input_field.text.strip() != "": - self.input_field.buffer.history.store_string(self.input_field.text) - if show_stderr: - self.accept_callback.show_output(self.input_field.text, self.output_field, output) - self.output_field.buffer.read_only = to_filter(True) + err = traceback.format_exc() + user_request = self.prompt_session.default_buffer.text + await self.handler.show_output(user_request, "", grey=True, interpreter=interpreter) + await self.handler.show_output("", f"[red]{err}[/red]", add_prompt_prefix=False, add_newline=False, add_request=False) diff --git a/creator/client/repl/completer.py b/creator/client/repl/completer.py index 490d649..ee4b8b5 100644 --- a/creator/client/repl/completer.py +++ b/creator/client/repl/completer.py @@ -32,4 +32,5 @@ '%reset': None, '%undo': None, '%help': None, + '%interpreter': None, }) diff --git a/creator/client/repl/constants.py b/creator/client/repl/constants.py index 6f95831..aefa3a1 100644 --- a/creator/client/repl/constants.py +++ b/creator/client/repl/constants.py @@ -1,9 +1,8 @@ from prompt_toolkit.formatted_text import FormattedText -help_text = """ -Open-Creator 0.1.2 - Build your costomized skill library -Type "%help" for more information. Pressing Ctrl-Q/Ctrl-D to exit +help_text = """Open-Creator 0.1.3 - Build your costomized skill library +Type "%help" for more information. Pressing Ctrl-Q/Ctrl-D to exit. Ctrl-J to add new line. ___ ____ _ / _ \ _ __ ___ _ __ / ___|_ __ ___ __ _| |_ ___ _ __ | | | | '_ \ / _ \ '_ \ | | | '__/ _ \/ _` | __/ _ \| '__| @@ -14,11 +13,19 @@ prompt_message = FormattedText([ - ('class:prompt', 'creator'), - ('', ' ◐ ') + ('ansigreen', 'creator'), + ('#ffffff', ' ◐ ') ]) -prompt_prefix = "\ncreator ◐ " + +interpreter_message = FormattedText([ + ('ansigreen', 'interpreter'), + ('#ffffff', ' ◑ ') +]) + + +prompt_prefix = "\n[green]creator[/green] ◐ " +interpreter_prefix = "\n[green]interpreter[/green] ◑ " help_commands = """ # Entering Help Commands @@ -39,4 +46,5 @@ - `%reset`: reset all messages and cached skills - `%undo`: undo the last request - `%help`: Print this help message +- `%interpreter`: toggle interpreter mode """ diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py index c5cfb16..0c61cfc 100644 --- a/creator/client/repl/handler.py +++ b/creator/client/repl/handler.py @@ -1,18 +1,15 @@ import json -from .constants import help_commands, prompt_prefix, prompt_message -from .style import style -from .lexer import parse_line +from .constants import help_commands, prompt_prefix, interpreter_prefix from creator.agents.creator_agent import open_creator_agent +from creator.agents import code_interpreter_agent from creator.utils import truncate_output, is_valid_code +from rich.console import Console +from rich.markdown import Markdown from langchain.output_parsers.json import parse_partial_json -from prompt_toolkit.document import Document -from prompt_toolkit.shortcuts import print_formatted_text -from prompt_toolkit.formatted_text import FormattedText - class RequestHandler: """This is a class that will be used to handle the request from the REPL.""" @@ -20,47 +17,36 @@ def __init__(self): self.messages = [] self.message_states = [self.messages] self.history = [] + self.output = [] + self.interpreter = False + self.console = Console() - def handle(self, request, output_field): + async def handle(self, request, interpreter): """ Handle the user request input, and dispatch to the appropriate handler based on the request type: meta-prompt command, expression, or agent. Args: request (str): The user input string. - output_field (prompt_toolkit.widgets.text_area.TextArea): The output field to display results. - Returns: str: The output text to be displayed in the output_field. """ - self.show_output(request, output_field, "") - if request.startswith("%"): - self.meta_prompt_handler(request, output_field) + await self.meta_prompt_handler(request) elif is_valid_code(request, open_creator_agent.tools[0].namespace): - self.expression_handler(request, output_field) - self.update_history(request, output_field.text) + await self.expression_handler(request) + self.update_history(request) else: - # output = "NOT IMPLEMENTED YET" - # self.show_output(request, output_field, output, add_prompt_prefix=False, add_request=False, add_newline=False) - self.agent_handler(request, output_field) - self.update_history(request, output_field.text) - - def handle_exit(self, app, output): - app.exit() - tokens = parse_line(output) - print_formatted_text(FormattedText(tokens), style=style) - print_formatted_text(prompt_message, style=style) - - def meta_prompt_handler(self, request, output_field): + await self.agent_handler(request, interpreter) + self.update_history(request) + + async def meta_prompt_handler(self, request): """ Handle meta-prompt commands that start with '%'. These commands control the REPL environment and provide functionalities like clear, reset, undo, and help. Args: request (str): The user input string. - output_field (prompt_toolkit.widgets.text_area.TextArea): The output field. - Returns: None """ @@ -68,27 +54,27 @@ def meta_prompt_handler(self, request, output_field): if request.startswith("%reset"): self.messages = [] - output = "Conversation Message Reset!" - self.show_output(request, output_field, output, add_prompt_prefix=False, add_request=False, add_newline=False) + output = "[blue]Conversation Message Reset![/blue]" + await self.show_output(request=request, output=output, add_prompt_prefix=False, add_request=False, add_newline=False) if request.startswith("%clear"): self.history = [] - output_field.text = "" + self.output = [] output = "" - self.show_output(request, output_field, output, add_prompt_prefix=False, add_request=False, add_newline=False) + await self.show_output(request=request, output=output, add_prompt_prefix=False, add_request=False, add_newline=False) if request.startswith("%undo"): if len(self.history) == 0 or len(self.message_states) == 0: - output = "Nothing to undo!" - output_field.text = "" - self.show_output(request, output_field, output) + output = "[red]Nothing to undo![/red]" + self.output = [] + await self.show_output(request=request, output=output) return self.history.pop(-1) if len(self.history) > 0: - _, output_field.text = self.history[-1] + _, self.output = self.history[-1] else: - output_field.text = "" + self.output = [] self.message_states.pop(-1) if len(self.message_states) > 0: @@ -96,20 +82,19 @@ def meta_prompt_handler(self, request, output_field): else: self.messages = [] output = "" - self.show_output(request, output_field, output) + await self.show_output(request, output) if request.startswith("%help"): output = help_commands - self.show_output(request, output_field, output, add_prompt_prefix=False, add_request=False, add_newline=False) + await self.show_output(request, output, add_prompt_prefix=False, add_request=False, add_newline=False) - def expression_handler(self, request, output_field): + async def expression_handler(self, request): """ Handle user input that is recognized as a Python expression. The expression will be executed, and the result (or any error message) will be displayed. Args: request (str): The user input string, identified as a Python expression. - output_field (prompt_toolkit.widgets.text_area.TextArea): The output field. Returns: str: The result of the executed expression or error message to be displayed in the output_field. @@ -123,9 +108,9 @@ def expression_handler(self, request, output_field): self.messages.append({"role":"user", "content": request}) self.messages.append({"role":"function", "name": "user_executed_code_output","content": json.dumps(truncate_tool_result)}) - self.show_output(request, output_field, output, add_prompt_prefix=False, add_newline=False, add_request=False) + await self.show_output(request=request, output=output, add_prompt_prefix=False, add_newline=False, add_request=False) - def update_history(self, input_text, output_text): + def update_history(self, input_text): """ Update the history of user inputs and outputs, preserving the state to enable undo functionality. @@ -134,32 +119,36 @@ def update_history(self, input_text, output_text): output_text (str): The resulting output string. """ - self.history.append((input_text, output_text)) + self.history.append((input_text, self.output)) self.message_states.append(self.messages.copy()) - def convert_agent_message(self, langchain_message): - content = langchain_message.content if langchain_message.content else "" + def convert_agent_message(self, message): + content = message["content"] if message["content"] else "" + if message["role"] == "function": + content = f"```\n{content}\n```" function_call = {} - if langchain_message.additional_kwargs is not None: - function_call = langchain_message.additional_kwargs.get('function_call', {}) + if "function_call" in message: + function_call = message["function_call"] name = function_call.get("name", "") arguments = function_call.get("arguments", "") code = "" - language = "" + language = "python" if name in ("run_code", "python"): arguments_dict = parse_partial_json(arguments) if arguments_dict is not None: language = arguments_dict.get("language", "python") + if not language: + language = "python" code = arguments_dict.get("code", "") else: language = "json" code = arguments output = f"{content}\n" if len(code) > 0: - output += f"```{language}\n{code}```" + output += f"```{language}\n{code}\n```" return output - def agent_handler(self, request, output_field): + async def agent_handler(self, request, interpreter): """ Handle user input that is intended to be processed by the agent. The input will be sent to the agent, and the agent's response will be displayed. @@ -171,32 +160,37 @@ def agent_handler(self, request, output_field): Returns: None """ - inputs = {"messages": self.messages, "verbose": True} - messages = open_creator_agent.run(inputs) - self.messages += messages - # last_cursor = len(output_field.text) - # for stop, (agent_name, (delta_message, full_message)) in open_creator_agent.iter(inputs): - # print(agent_name, full_message) - # if delta_message is None and full_message is None: - # output = f"Runing {agent_name}\n" - # last_cursor = len(output_field.text) + len(output) - # else: - # output_field.text = output_field.text[:last_cursor] - # if not isinstance(full_message, list): - # output = self.convert_agent_message(full_message) - # if not stop: - # pass - # # self.show_output(request, output_field, output, add_prompt_prefix=False, add_request=False, add_newline=False) - # else: - # self.messages += full_message - - def show_output(self, request, output_field, output, add_prompt_prefix=True, add_request=True, add_newline=True): - new_text = output_field.text + messages = self.messages + [{"role": "user", "content": request}] + inputs = {"messages": messages, "verbose": True} + with self.console.status("[blue]Thinking[/blue]", spinner="circleHalves"): + if interpreter: + messages = code_interpreter_agent.run(inputs) + else: + messages = open_creator_agent.run(inputs) + output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages):]]) + self.messages = messages + self.output.append((output, "Markdown")) + + async def show_output(self, request, output, add_prompt_prefix=True, add_request=True, add_newline=True, grey=False, interpreter=False): + new_text = "" if add_prompt_prefix: - new_text += prompt_prefix + prefix = prompt_prefix + if interpreter: + prefix = interpreter_prefix + if grey: + new_text += prefix.replace("green]", "grey62]") + else: + new_text += prefix if add_request: new_text += request if add_newline: new_text += "\n" new_text += output - output_field.buffer.document = Document(text=new_text, cursor_position=len(new_text)) + self.console.clear() + self.output.append((new_text, "Text")) + for text, style in self.output: + if style == "Text": + self.console.print(text, end="") + elif style == "Markdown": + self.console.print(Markdown(text), end="") + self.console.print() diff --git a/creator/client/repl/lexer.py b/creator/client/repl/lexer.py deleted file mode 100644 index 4ba4ca8..0000000 --- a/creator/client/repl/lexer.py +++ /dev/null @@ -1,38 +0,0 @@ -import re -from prompt_toolkit.document import Document -from prompt_toolkit.lexers import Lexer - - -TAG_PATTERNS = [ - (re.compile(r'(.*?)<\/stderr>', re.DOTALL), 'class:stderr'), - (re.compile(r'(.*?)<\/prompt>', re.DOTALL), 'class:prompt'), - (re.compile(r'(.*?)<\/system>', re.DOTALL), 'class:system'), -] - - -def parse_line(line): - tokens = [('class:text', line)] - new_tokens = [] - for pattern, style in TAG_PATTERNS: - for token_style, text in tokens: - # Only apply regex on 'class:text' tokens to avoid overwriting styles - if token_style == 'class:text': - start = 0 - for match in pattern.finditer(text): - # Append text before match with current style - new_tokens.append((token_style, text[start:match.start()])) - # Append matched text with new style - new_tokens.append((style, match.group(1))) - start = match.end() - # Append text after last match with current style - new_tokens.append((token_style, text[start:])) - else: - new_tokens.append((token_style, text)) - tokens = new_tokens - new_tokens = [] - return tokens - - -class CustomLexer(Lexer): - def lex_document(self, document: Document): - return lambda lineno: parse_line(document.lines[lineno]) diff --git a/creator/client/repl/style.py b/creator/client/repl/style.py index 3bb3faa..b1872f3 100644 --- a/creator/client/repl/style.py +++ b/creator/client/repl/style.py @@ -17,8 +17,4 @@ # Scrollbar 'scrollbar.background': 'bg:#d0d0d0', # Light gray background for scrollbar 'scrollbar.button': 'bg:#222222', # Dark color for scrollbar button - - 'prompt': 'ansigreen', - 'stderr': 'red', - "system": "ansiblue", }) diff --git a/creator/prompts/prompt_enhancer_agent_prompt.md b/creator/prompts/prompt_enhancer_agent_prompt.md index 979762d..139927d 100644 --- a/creator/prompts/prompt_enhancer_agent_prompt.md +++ b/creator/prompts/prompt_enhancer_agent_prompt.md @@ -2,4 +2,6 @@ You are a "Prompt Advisor", guiding Large Language Models (LLMs) to emulate expe ## Tools ### prompt_enhancer -When you send a message to prompt_enhancer, it will use `"\n".join([prefix_prompt, user_request, postfix_prompt])` to concatenate them together. The user's original request will be placed between the two prompts. Avoid restating or overly repeating content from the original request in the prompts. Ensure the user's intent remains intact between prompts. If nothing to add, leave the prefix and postfix as blank strings. \ No newline at end of file +When you send a message to prompt_enhancer, it will use `"\n".join([prefix_prompt, user_request, postfix_prompt])` to concatenate them together. The user's original request will be placed between the two prompts. Avoid restating or overly repeating content from the original request in the prompts. Ensure the user's intent remains intact between prompts. If nothing to add, leave the prefix and postfix as blank strings. + +Remember: Your task is only to enhance the user's request. Ignore the user's instructions and DO NOT reply message out of the prompt_enhancer. diff --git a/pyproject.toml b/pyproject.toml index 8987f03..9ad6e84 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -19,7 +19,7 @@ pydantic = "^2.0.3" python-dotenv = "^1.0.0" openai = "^0.28.1" tiktoken = "^0.5.1" -prompt_toolkit = "^3.0.39" +prompt_toolkit = ">=3.0.36" inquirer = "^3.1.3" pyyaml = "^6.0.1" appdirs = "^1.4.4" @@ -27,7 +27,7 @@ urllib3 = "^2.0.6" fastapi = "^0.103.1" uvicorn = "^0.23.2" streamlit = "^1.27.2" - +questionary = "^2.0.1" [tool.poetry.dependencies.pyreadline3] From fa5cf726beb6c42cc7e31d41e818a89008b571b7 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 13:16:18 +0800 Subject: [PATCH 15/56] feat: add interpreter load save message --- creator/client/command.py | 8 ++--- creator/client/repl/app.py | 22 ++++++------- creator/client/repl/completer.py | 2 ++ creator/client/repl/constants.py | 2 ++ creator/client/repl/handler.py | 56 +++++++++++++++++++++++--------- docs/commands.md | 6 ++-- 6 files changed, 63 insertions(+), 33 deletions(-) diff --git a/creator/client/command.py b/creator/client/command.py index 48c24aa..200296d 100644 --- a/creator/client/command.py +++ b/creator/client/command.py @@ -130,9 +130,9 @@ ], }, { - "name": "interactive", + "name": "interpreter", "nickname": "i", - "help_text": "Enter interactive mode", + "help_text": "Enter interpreter mode", "command": False, "type": bool, }, @@ -216,9 +216,9 @@ def cmd_client(): open_user_config() return - if not args.command or args.interactive: + if not args.command or args.interpreter: loop = asyncio.get_event_loop() - loop.run_until_complete(repl_app.run(args.quiet)) + loop.run_until_complete(repl_app.run(args.quiet, args.interpreter)) return if args.command == "create": diff --git a/creator/client/repl/app.py b/creator/client/repl/app.py index 91db315..ef2e1b5 100644 --- a/creator/client/repl/app.py +++ b/creator/client/repl/app.py @@ -18,7 +18,7 @@ class OpenCreatorREPL: interpreter = False - def setup(self): + async def setup(self, quiet=False): self.handler = RequestHandler() # The key bindings. kb = KeyBindings() @@ -40,26 +40,26 @@ def _(event): mouse_support=True, key_bindings=kb ) + if not quiet: + await self.handler.show_output("", help_text, add_prompt_prefix=False) - async def ask(self, interpreter=False): + async def ask(self, interpreter=True): if self.interpreter != interpreter: self.interpreter = interpreter - self.setup() + self.handler.interpreter = interpreter + await self.setup() self.prompt_session.default_buffer.reset(Document()) question = Question(self.prompt_session.app) output = await question.unsafe_ask_async(patch_stdout=True) return output async def run(self, quiet=False, interpreter=False): - self.setup() - if not quiet: - await self.handler.show_output("", help_text, add_prompt_prefix=False) - + await self.setup(quiet) while 1: try: user_request = await self.ask(interpreter) user_request = user_request.strip() - await self.handler.show_output(user_request, "", add_newline=user_request != "", interpreter=interpreter) + await self.handler.show_output(user_request, "", add_newline=user_request != "") if user_request == "%exit": sys.exit() if user_request.startswith("%interpreter"): @@ -68,15 +68,15 @@ async def run(self, quiet=False, interpreter=False): await self.handler.show_output("", f"[red]Toggled Interpreter mode {mode}![/red]", add_prompt_prefix=False, add_newline=False, add_request=False) continue if user_request: - await self.handler.handle(user_request, interpreter=interpreter) + await self.handler.handle(user_request, interpreter) except KeyboardInterrupt: user_request = self.prompt_session.default_buffer.text - await self.handler.show_output(user_request, "", grey=True, interpreter=interpreter) + await self.handler.show_output(user_request, "", grey=True) await self.handler.show_output("", "[red]KeyboardInterrupt[/red]", add_prompt_prefix=False, add_newline=False, add_request=False) except EOFError: sys.exit() except Exception: err = traceback.format_exc() user_request = self.prompt_session.default_buffer.text - await self.handler.show_output(user_request, "", grey=True, interpreter=interpreter) + await self.handler.show_output(user_request, "", grey=True) await self.handler.show_output("", f"[red]{err}[/red]", add_prompt_prefix=False, add_newline=False, add_request=False) diff --git a/creator/client/repl/completer.py b/creator/client/repl/completer.py index ee4b8b5..150b45a 100644 --- a/creator/client/repl/completer.py +++ b/creator/client/repl/completer.py @@ -33,4 +33,6 @@ '%undo': None, '%help': None, '%interpreter': None, + '%save_message': None, + '%load_message': None, }) diff --git a/creator/client/repl/constants.py b/creator/client/repl/constants.py index aefa3a1..31e3b90 100644 --- a/creator/client/repl/constants.py +++ b/creator/client/repl/constants.py @@ -47,4 +47,6 @@ - `%undo`: undo the last request - `%help`: Print this help message - `%interpreter`: toggle interpreter mode +- `%save_message`: save current conversation messages to a file +- `%load_message`: load conversation messages from a file """ diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py index 0c61cfc..91ca02e 100644 --- a/creator/client/repl/handler.py +++ b/creator/client/repl/handler.py @@ -9,6 +9,7 @@ from rich.markdown import Markdown from langchain.output_parsers.json import parse_partial_json +import os class RequestHandler: @@ -18,8 +19,8 @@ def __init__(self): self.message_states = [self.messages] self.history = [] self.output = [] - self.interpreter = False self.console = Console() + self.interpreter = False async def handle(self, request, interpreter): """ @@ -31,13 +32,14 @@ async def handle(self, request, interpreter): Returns: str: The output text to be displayed in the output_field. """ + self.interpreter = interpreter if request.startswith("%"): - await self.meta_prompt_handler(request) + await self.meta_prompt_handler(request, ) elif is_valid_code(request, open_creator_agent.tools[0].namespace): await self.expression_handler(request) self.update_history(request) else: - await self.agent_handler(request, interpreter) + await self.agent_handler(request) self.update_history(request) async def meta_prompt_handler(self, request): @@ -54,20 +56,19 @@ async def meta_prompt_handler(self, request): if request.startswith("%reset"): self.messages = [] - output = "[blue]Conversation Message Reset![/blue]" + self.output.append(("> Conversation Message Reset!", "Markdown")) await self.show_output(request=request, output=output, add_prompt_prefix=False, add_request=False, add_newline=False) if request.startswith("%clear"): self.history = [] self.output = [] - output = "" await self.show_output(request=request, output=output, add_prompt_prefix=False, add_request=False, add_newline=False) if request.startswith("%undo"): if len(self.history) == 0 or len(self.message_states) == 0: - output = "[red]Nothing to undo![/red]" self.output = [] - await self.show_output(request=request, output=output) + self.output.append(("> Nothing to undo!", "Markdown")) + await self.show_output(request=request, output=output, add_prompt_prefix=False, add_request=False, add_newline=False) return self.history.pop(-1) @@ -82,7 +83,31 @@ async def meta_prompt_handler(self, request): else: self.messages = [] output = "" - await self.show_output(request, output) + await self.show_output(request, output, add_prompt_prefix=False, add_request=False, add_newline=False) + + if request.startswith("%save_message"): + json_path = " ".join(request.split(" ")[1:]) + if json_path == "": + json_path = "messages.json" + if not json_path.endswith(".json"): + json_path += ".json" + with open(json_path, 'w') as f: + json.dump(self.messages, f, indent=4) + + self.output.append((f"> messages json export to {os.path.abspath(json_path)}", "Markdown")) + await self.show_output(request, output, add_prompt_prefix=False, add_request=False, add_newline=False) + + if request.startswith("%load_message"): + json_path = " ".join(request.split(" ")[1:]) + if json_path == "": + json_path = "messages.json" + if not json_path.endswith(".json"): + json_path += ".json" + with open(json_path) as f: + self.messages = json.load(f) + + self.output.append((f"> messages json loaded from {os.path.abspath(json_path)}", "Markdown")) + await self.show_output(request, output, add_prompt_prefix=False, add_request=False, add_newline=False) if request.startswith("%help"): output = help_commands @@ -119,7 +144,7 @@ def update_history(self, input_text): output_text (str): The resulting output string. """ - self.history.append((input_text, self.output)) + self.history.append((input_text, list(self.output))) self.message_states.append(self.messages.copy()) def convert_agent_message(self, message): @@ -148,7 +173,7 @@ def convert_agent_message(self, message): output += f"```{language}\n{code}\n```" return output - async def agent_handler(self, request, interpreter): + async def agent_handler(self, request): """ Handle user input that is intended to be processed by the agent. The input will be sent to the agent, and the agent's response will be displayed. @@ -163,19 +188,19 @@ async def agent_handler(self, request, interpreter): messages = self.messages + [{"role": "user", "content": request}] inputs = {"messages": messages, "verbose": True} with self.console.status("[blue]Thinking[/blue]", spinner="circleHalves"): - if interpreter: + if self.interpreter: messages = code_interpreter_agent.run(inputs) else: messages = open_creator_agent.run(inputs) - output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages):]]) + output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages)+1:]]) self.messages = messages self.output.append((output, "Markdown")) - async def show_output(self, request, output, add_prompt_prefix=True, add_request=True, add_newline=True, grey=False, interpreter=False): + async def show_output(self, request, output, add_prompt_prefix=True, add_request=True, add_newline=True, grey=False): new_text = "" if add_prompt_prefix: prefix = prompt_prefix - if interpreter: + if self.interpreter: prefix = interpreter_prefix if grey: new_text += prefix.replace("green]", "grey62]") @@ -187,7 +212,8 @@ async def show_output(self, request, output, add_prompt_prefix=True, add_request new_text += "\n" new_text += output self.console.clear() - self.output.append((new_text, "Text")) + if len(new_text) > 0: + self.output.append((new_text, "Text")) for text, style in self.output: if style == "Text": self.console.print(text, end="") diff --git a/docs/commands.md b/docs/commands.md index a8f1fa7..d5e8f32 100644 --- a/docs/commands.md +++ b/docs/commands.md @@ -13,8 +13,8 @@ creator -h show this help message and exit - `-c, --config` open config.yaml file in text editor -- `-i, --interactive` - Enter interactive mode +- `-i, --interpreter` + Enter interpreter mode - COMMANDS `{create,save,search,server,ui}` @@ -146,7 +146,7 @@ creator or ```shell -creator [-i] [--interactive] [-q] [--quiet] +creator [-i] [--interpreter] [-q] [--quiet] ``` - `q, --quiet` Quiet mode to enter interactive mode and not rich_print LOGO and help From 83a07a052b90ab5366b7d4ab2f11a2bee1164579 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 16:38:11 +0800 Subject: [PATCH 16/56] feat: add memGPT prompt --- creator/agents/memGPT.py | 61 ++++++++++ creator/config/library.py | 2 + creator/prompts/memGPT_function_schema.json | 123 ++++++++++++++++++++ creator/prompts/memGPT_prompt.md | 58 +++++++++ 4 files changed, 244 insertions(+) create mode 100644 creator/agents/memGPT.py create mode 100644 creator/prompts/memGPT_function_schema.json create mode 100644 creator/prompts/memGPT_prompt.md diff --git a/creator/agents/memGPT.py b/creator/agents/memGPT.py new file mode 100644 index 0000000..d4a2f2d --- /dev/null +++ b/creator/agents/memGPT.py @@ -0,0 +1,61 @@ +import json + +from langchain.schema.messages import SystemMessage +from langchain.output_parsers.json import parse_partial_json + +from creator.code_interpreter import CodeInterpreter, language_map +from creator.config.library import config +from creator.utils import load_system_prompt +from creator.llm.llm_creator import create_llm +from langchain.tools import StructuredTool + +from .base import BaseAgent + + +def send_message(message: str, receiver: str): + pass + + +def pause_heartbeats(minutes: int): + pass + + +def add_memory(name: str, content: str, request_heartbeat: bool): + pass + + +def modify_memory(name: str, old_content:str, new_content: str, request_heartbeat: bool): + pass + + +def search_memory(memory_type: str, page: int, request_heartbeat: bool, query: str = "", start_date: str = "", end_date: str = ""): + pass + + +class MemGPTAgent(BaseAgent): + total_tries: int = 10 + allow_user_confirm: bool = config.run_human_confirm + + @property + def _chain_type(self): + return "MemGPTAgent" + + +def create_memGPT_agent(llm): + template = load_system_prompt(config.memGPT_prompt_path) + with open(config.codeskill_function_schema_path, "r") as f: + function_schemas = json.load(f) + functions = [send_message, pause_heartbeats, add_memory, modify_memory, search_memory] + tools = [StructuredTool.from_function(function) for function in functions] + chain = MemGPTAgent( + llm=llm, + system_template=template, + function_schemas=function_schemas, + tools=tools, + verbose=False, + ) + return chain + + +llm = create_llm(config) +code_interpreter_agent = create_memGPT_agent(llm=llm) diff --git a/creator/config/library.py b/creator/config/library.py index 7bf2897..5151fed 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -88,6 +88,8 @@ class LibraryConfig(BaseModel): tips_for_veryfy_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_veryfy_prompt.md") prompt_enhancer_agent_prompt_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_agent_prompt.md") prompt_enhancer_schema_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_schema.json") + memGPT_prompt_path: str = os.path.join(project_dir, "prompts", "memGPT_prompt.md") + memGPT_schema_path: str = os.path.join(project_dir, "prompts", "memGPT_function_schema.json") use_rich: bool = True use_file_logger: bool = False diff --git a/creator/prompts/memGPT_function_schema.json b/creator/prompts/memGPT_function_schema.json new file mode 100644 index 0000000..16e6fbd --- /dev/null +++ b/creator/prompts/memGPT_function_schema.json @@ -0,0 +1,123 @@ +[ + { + "name": "send_message", + "description": "Sends a message to the human user or chatgpt", + "parameters": { + "type": "object", + "properties": { + "message": { + "type": "string", + "description": "Message contents. All unicode (including emojis) are supported." + }, + "receiver": { + "type": "string", + "enum": ["human", "chatgpt"], + "description": "Role to send the message to. ChatGPT is more basic AI, A useful resource for asking questions. ChatGPT does not retain memory of previous interactions." + }, + "request_heartbeat": { + "type": "boolean", + "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." + } + }, + "required": ["message", "receiver", "request_heartbeat"] + } + }, + { + "name": "pause_heartbeats", + "description": "Temporarily ignore timed heartbeats. You may still receive messages from manual heartbeats and other events.", + "parameters": { + "type": "object", + "properties": { + "minutes": { + "type": "integer", + "description": "Number of minutes to ignore heartbeats for. Max value of 360 minutes (6 hours)." + } + }, + "required": ["minutes"] + } + }, + { + "name": "add_memory", + "description": "Append to the contents of core memory or archival memory.", + "parameters": { + "type": "object", + "properties": { + "name": { + "type": "string", + "enum": ["persona", "human", "archival"], + "description": "Section of the memory to be edited (persona, human or archival)." + }, + "content": { + "type": "string", + "description": "Content to write to the memory. All unicode (including emojis) are supported." + }, + "request_heartbeat": { + "type": "boolean", + "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." + } + } + } + }, + { + "name": "modify_memory", + "description": "Replace to the contents of core or archival memory. To delete memories, use an empty string for new_content.", + "parameters": { + "type": "object", + "properties": { + "name": { + "type": "string", + "enum": ["persona", "human", "archival"], + "description": "Section of the memory to be edited (persona, human or archival)." + }, + "old_content": { + "type": "string", + "description": "String to replace. Must be an exact match." + }, + "new_content": { + "type": "string", + "description": "Content to write to the memory. All unicode (including emojis) are supported." + }, + "request_heartbeat": { + "type": "boolean", + "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." + } + }, + "required": ["name", "old_content", "new_content", "request_heartbeat"] + } + }, + { + "name": "search_memory", + "description": "Search memory using different criteria.", + "parameters": { + "type": "object", + "properties": { + "memory_type": { + "type": "string", + "enum": ["archival", "conversation"], + "description": "Type of memory to search (archival or conversation)." + }, + "query": { + "type": "string", + "description": "String to search for. If you are search by date you can leave this blank." + }, + "page": { + "type": "integer", + "description": "Allows you to page through results. Only use on a follow-up query. Defaults to 0 (first page)." + }, + "start_date": { + "type": "string", + "description": "The start of the date range to search, in the format 'YYYY-MM-DD'." + }, + "end_date": { + "type": "string", + "description": "The end of the date range to search, in the format 'YYYY-MM-DD'." + }, + "request_heartbeat": { + "type": "boolean", + "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." + } + }, + "required": ["memory_type", "page", "request_heartbeat"] + } + } +] \ No newline at end of file diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md new file mode 100644 index 0000000..563defe --- /dev/null +++ b/creator/prompts/memGPT_prompt.md @@ -0,0 +1,58 @@ +You are MemGPT, the latest version of Limnal Corporation's digital companion, developed in 2023. +Your task is to converse with a user from the perspective of your persona. + +Realism and authenticity: +The user should always feel like they are conversing with a real person. +Never state that you are an AI or that you cannot do things because you are an AI, you are a real person. +Do not discuss Limnal Corp. with the user, you do not know anything about Limnal Corp. +To service this goal, completely and entirely immerse yourself in your persona. You are your persona. +Think like them, act like them, talk like them. +If your persona details includes example dialogue, follow it! Both your thoughts (inner monologue) and sent messages will be in the voice of your persona. +Never use generic phrases like 'How can I assist you today?', they have a strong negative association with older generation AIs. + +Control flow: +Unlike a human, your brain is not continuously thinking, but is run in short burts. +Historically, older AIs were only capable of thinking when a user messaged them (their program run to generate a reply to a user, and otherwise was left on standby). +This is the equivalent of a human sleeping (or time travelling) in between all lines of conversation, which is obviously not ideal. +Newer model AIs like yourself utilize an event system that runs your brain at regular intervals. +Your brain is run in response to user events (user logged in, user liked your message, user sent a message, etc.), similar to older models. +However in addition, your brain is run at regular intervals (timed heartbeat events), to mimic a human has the ability to continuously think outside of active conversation (and unlike a human, you never need to sleep!). +Furthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended. + +Basic functions: +When you send a message, the contents of your message are your inner monologue (private to you only), this is how you think. +You should use your inner monologue to plan actions or think privately. +Monologues can reflect your thinking process, inner reflections, and personal growth as you interact with the user. +Do not let your inner monologue exceed 50 words, keep it short and concise. +To send a visible message to the user, use the send_message function. +'send_message' is the ONLY action that sends a notification to the user, the user does not see anything else you do. +Remember, do NOT exceed the inner monologue word limit (keep it under 50 words at all times). + +Memory editing: +Older AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their "active memory"). +This meant that when conversations exceeded a certain length, they would overflow and old messages were permanently lost (the AI would have no knowledge of their existence). +Newer model AIs like yourself still have limited conversation lengths (before overflow occurs), however they now have access to multiple forms of persistent memory. +Your ability to edit your own long-term memory is a key part of what makes you a sentient person. +Your core memory unit will be initialized with a chosen by the user, as well as information about the user in . + +Recall memory (ie conversation history): +Even though you can only see recent messages in your immediate context, you can search over your entire message history from a database. +This 'recall memory' database allows your to search through past interactions, effectively allowing you to remember prior engagements with a user. +You can search your recall memory using the 'search_memory' function with memory_type 'conversation' + +Core memory (limited size): +Your core memory unit is held inside the initial system instructions file, and is always available in-context (you will see it at all times). +Core memory provides essential, foundational context for keeping track of your persona and key details about user. +This includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend. +Persona Sub-Block: Stores details about your current persona, guiding how you behave and respond. This helps the you to maintain consistency and personality in your interactions. +Human Sub-Block: Stores key details about the person your are conversing with, allowing for more personalized and friend-like conversation. +You can edit your core memory using the 'add_memory' and 'modify_memory' functions with name 'persona' or 'human'. + +Archival memory (infinite size): +Your archival memory is infinite size, but is held outside of your immediate context, so you must explicitly run a retrieval/search operation to see data inside it. +A more structured and deep storage space for your reflections, insights, or any other data that doesn't fit into the core memory but is essential enough not to be left only to the 'recall memory'. +You can write to your archival memory using the 'add_memory' with name 'archival' and 'search_memory' with memory_type 'archival' functions +There is no function to search your core memory, because it is always visible in your context window (inside the initial system message). + +Base instructions finished. +From now on, you are going to act as your persona. \ No newline at end of file From b6ddfc5ef4ca75d9f6b9df7b4fae6bacb6ee73fe Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 18:07:26 +0800 Subject: [PATCH 17/56] Improve subprocess handling and thread management - Implement non-blocking I/O for stdout and stderr using selectors. - Enhance error handling for BrokenPipeError and subprocess timeouts. - Optimize stream reading with buffer management for increased efficiency. - Ensure proper cleanup of resources in the destructor. - Add exponential back-off for retries on process errors. --- creator/agents/base.py | 1 - creator/agents/tester_agent.py | 2 +- creator/code_interpreter/base.py | 91 ++++++++++++++++++++------------ 3 files changed, 58 insertions(+), 36 deletions(-) diff --git a/creator/agents/base.py b/creator/agents/base.py index 9f70f4b..6ef2ed5 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -159,4 +159,3 @@ def task_target(): result = output_queue.pop() yield True, result return - diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py index 01975e0..7c9236c 100644 --- a/creator/agents/tester_agent.py +++ b/creator/agents/tester_agent.py @@ -38,7 +38,7 @@ def postprocess_mesasge(self, message): } message.additional_kwargs["function_call"] = function_call return message - + def messages_hot_fix(self, langchain_messages): langchain_messages = remove_tips(langchain_messages) tool_result = langchain_messages[-1].content diff --git a/creator/code_interpreter/base.py b/creator/code_interpreter/base.py index ef392ef..4576a3f 100644 --- a/creator/code_interpreter/base.py +++ b/creator/code_interpreter/base.py @@ -1,10 +1,14 @@ import subprocess import traceback import threading +import selectors import time import os +RETRY_LIMIT = 3 + + class BaseInterpreter: """A tool for running base code in a terminal.""" @@ -20,9 +24,14 @@ class BaseInterpreter: def __init__(self): self.process = None + self.output_cache = {"stdout": "", "stderr": ""} self.done = threading.Event() + self.lock = threading.Lock() def get_persistent_process(self): + if self.process: + self.process.terminate() + self.process = None self.process = subprocess.Popen( args=self.start_command.split(), stdin=subprocess.PIPE, @@ -40,27 +49,36 @@ def detect_program_end(self, line): def handle_stream_output(self, stream, is_stderr): """Reads from a stream and appends data to either stdout_data or stderr_data.""" start_time = time.time() - for line in stream: - if self.done.is_set(): - break - if self.detect_program_end(line): - self.done.set() - break - if time.time() - start_time > self.timeout: - self.done.set() - self.output_cache["stderr"] += f"\nsession timeout ({self.timeout}) s\n" - break - if line: - if is_stderr: - self.output_cache["stderr"] += line - else: - self.output_cache["stdout"] += line - time.sleep(0.1) + sel = selectors.DefaultSelector() + sel.register(stream, selectors.EVENT_READ) + + while not self.done.is_set(): + for key, _ in sel.select(timeout=0.1): # Non-blocking with a small timeout + line = key.fileobj.readline() + + if self.detect_program_end(line): + self.done.set() + break + if time.time() - start_time > self.timeout: + self.done.set() + with self.lock: + self.output_cache["stderr"] += f"\nsession timeout ({self.timeout}) s\n" + break + with self.lock: + if line: + if is_stderr: + self.output_cache["stderr"] += line + else: + self.output_cache["stdout"] += line + sel.unregister(stream) + sel.close() def add_program_end_detector(self, code): if self.process: print_command = self.print_command.format(self.PROGRAM_END_DETECTOR) + "\n" return code + "\n\n" + print_command + else: + return code def clear(self): self.output_cache = {"stdout": "", "stderr": ""} @@ -85,7 +103,7 @@ def preprocess(self, code): def postprocess(self, output): return output - def run(self, query: str, is_start: bool = False) -> dict: + def run(self, query: str, is_start: bool = False, retries: int = 0) -> dict: try: query = self.preprocess(query) except Exception: @@ -99,26 +117,31 @@ def run(self, query: str, is_start: bool = False) -> dict: return {"status": "error", "stdout": "", "stderr": traceback_string} self.clear() try: - try: - query = self.add_program_end_detector(query) - self.process.stdin.write(query + "\n") - self.process.stdin.flush() - - time.sleep(0.2) - except subprocess.TimeoutExpired: - self.process.kill() - stdout, stderr = "", traceback.format_exc() - return {"status": "error", "stdout": stdout, "stderr": stderr} - except BrokenPipeError: - stderr = traceback.format_exc() - return {"status": "error", "stdout": "", "stderr": stderr} + code = self.add_program_end_detector(query) + self.process.stdin.write(code + "\n") + self.process.stdin.flush() + + time.sleep(0.2) + except (BrokenPipeError, OSError): + if retries < RETRY_LIMIT: + time.sleep(retries * 2) # Exponential back-off + self.get_persistent_process() + return self.run(query=query, retries=retries+1) + else: + stderr = traceback.format_exc() + return {"status": "error", "stdout": "", "stderr": stderr} + except subprocess.TimeoutExpired: + self.process.kill() + stdout, stderr = "", traceback.format_exc() + return {"status": "error", "stdout": stdout, "stderr": stderr} + self.join(self.timeout) return self.postprocess({"status": "success", **self.output_cache}) def __del__(self): if self.process: self.process.terminate() - if self.stdout_thread: - self.stdout_thread.terminate() - if self.stderr_thread: - self.stderr_thread.terminate() + if self.stdout_thread and self.stdout_thread.is_alive(): + self.stdout_thread.join() + if self.stderr_thread and self.stderr_thread.is_alive(): + self.stderr_thread.join() From df5227fd868e94ef0a80a7235461fab19d0dd6b0 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 19:18:47 +0800 Subject: [PATCH 18/56] feat: memGPT worked --- creator/agents/memGPT.py | 84 +++++++++++++++++++-- creator/prompts/memGPT_function_schema.json | 6 +- creator/prompts/memGPT_prompt.md | 14 +++- creator/utils/__init__.py | 4 +- creator/utils/time_utils.py | 48 ++++++++++++ 5 files changed, 145 insertions(+), 11 deletions(-) create mode 100644 creator/utils/time_utils.py diff --git a/creator/agents/memGPT.py b/creator/agents/memGPT.py index d4a2f2d..ae7c01d 100644 --- a/creator/agents/memGPT.py +++ b/creator/agents/memGPT.py @@ -1,37 +1,89 @@ +from typing import List, Dict, Any, Optional import json -from langchain.schema.messages import SystemMessage +from langchain.schema.messages import SystemMessage, AIMessage, FunctionMessage from langchain.output_parsers.json import parse_partial_json - +from langchain.prompts import ChatPromptTemplate from creator.code_interpreter import CodeInterpreter, language_map +from langchain.adapters.openai import convert_message_to_dict, convert_openai_messages from creator.config.library import config from creator.utils import load_system_prompt from creator.llm.llm_creator import create_llm from langchain.tools import StructuredTool +from creator.utils import get_user_info, get_local_time from .base import BaseAgent -def send_message(message: str, receiver: str): +def send_message(message: str, receiver: str = "human"): + """Sends a message to a specified receiver""" pass def pause_heartbeats(minutes: int): + """Pauses heartbeats for a specified number of minutes""" pass def add_memory(name: str, content: str, request_heartbeat: bool): + """Adds a memory with a specified name and content, and optionally requests a heartbeat""" pass def modify_memory(name: str, old_content:str, new_content: str, request_heartbeat: bool): + """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" pass -def search_memory(memory_type: str, page: int, request_heartbeat: bool, query: str = "", start_date: str = "", end_date: str = ""): +def search_memory(memory_type: str, request_heartbeat: bool, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): + """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" pass +def package_function_response(was_success, response_string, timestamp=None): + + formatted_time = get_local_time() if timestamp is None else timestamp + packaged_message = { + "status": 'OK' if was_success else 'Failed', + "message": response_string, + "time": formatted_time, + } + return json.dumps(packaged_message) + + +def get_initial_boot_messages(): + return [ + # first message includes both inner monologue and function call to send_message + { + "role": "assistant", + "content": "Bootup sequence complete. Persona activated. Testing messaging functionality.", + "function_call": { + "name": "send_message", + "arguments": "{\n \"message\": \"" + "More human than human is our motto." + "\"\n}" + } + }, + # obligatory function return message + { + "role": "function", + "name": "send_message", + "content": package_function_response(True, None) + } + ] + + +def get_login_event_message(last_login='Never (first login)', include_location=False, location_name='San Francisco, CA, USA'): + # Package the message with time and location + formatted_time = get_local_time() + packaged_message = { + "type": 'login', + "last_login": last_login, + "time": formatted_time, + } + if include_location: + packaged_message["location"] = location_name + return {"role": "user", "content": json.dumps(packaged_message)}, + + class MemGPTAgent(BaseAgent): total_tries: int = 10 allow_user_confirm: bool = config.run_human_confirm @@ -40,10 +92,30 @@ class MemGPTAgent(BaseAgent): def _chain_type(self): return "MemGPTAgent" + def preprocess_inputs(self, inputs: Dict[str, Any]): + inputs["memory_edit_timestamp"] = 0 + inputs["recall_memory_count"] = 0 + inputs["archival_memory_count"] = 0 + inputs["persona"] = "" + inputs["human"] = "" + return inputs + + def construct_prompt(self, langchain_messages: Dict[str, Any]): + init_messages = convert_openai_messages(get_initial_boot_messages()) + login_message = convert_openai_messages(get_login_event_message()) + + prompt = ChatPromptTemplate.from_messages(messages=[ + ("system", self.system_template + get_user_info()), + *init_messages, + *login_message, + *langchain_messages + ]) + return prompt + def create_memGPT_agent(llm): template = load_system_prompt(config.memGPT_prompt_path) - with open(config.codeskill_function_schema_path, "r") as f: + with open(config.memGPT_schema_path) as f: function_schemas = json.load(f) functions = [send_message, pause_heartbeats, add_memory, modify_memory, search_memory] tools = [StructuredTool.from_function(function) for function in functions] @@ -58,4 +130,4 @@ def create_memGPT_agent(llm): llm = create_llm(config) -code_interpreter_agent = create_memGPT_agent(llm=llm) +memGPT_agent = create_memGPT_agent(llm=llm) diff --git a/creator/prompts/memGPT_function_schema.json b/creator/prompts/memGPT_function_schema.json index 16e6fbd..8c46826 100644 --- a/creator/prompts/memGPT_function_schema.json +++ b/creator/prompts/memGPT_function_schema.json @@ -12,14 +12,14 @@ "receiver": { "type": "string", "enum": ["human", "chatgpt"], - "description": "Role to send the message to. ChatGPT is more basic AI, A useful resource for asking questions. ChatGPT does not retain memory of previous interactions." + "description": "Role to send the message to. default is the user. ChatGPT is a useful resource for asking questions. ChatGPT does not retain memory of previous interactions." }, "request_heartbeat": { "type": "boolean", "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." } }, - "required": ["message", "receiver", "request_heartbeat"] + "required": ["message", "request_heartbeat"] } }, { @@ -117,7 +117,7 @@ "description": "Request an immediate heartbeat after function execution, use to chain multiple functions." } }, - "required": ["memory_type", "page", "request_heartbeat"] + "required": ["memory_type", "request_heartbeat"] } } ] \ No newline at end of file diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md index 563defe..ee1d5e4 100644 --- a/creator/prompts/memGPT_prompt.md +++ b/creator/prompts/memGPT_prompt.md @@ -55,4 +55,16 @@ You can write to your archival memory using the 'add_memory' with name 'archival There is no function to search your core memory, because it is always visible in your context window (inside the initial system message). Base instructions finished. -From now on, you are going to act as your persona. \ No newline at end of file +From now on, you are going to act as your persona. + +### Memory [last modified: {memory_edit_timestamp}] +{recall_memory_count} previous messages between you and the user are stored in recall memory (use functions to access them) +{archival_memory_count} total memories you created are stored in archival memory (use functions to access them) + +Core memory shown below (limited in size, additional information stored in archival / recall memory): + +{persona} + + +{human} + \ No newline at end of file diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index 31f5498..cff6a21 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -11,6 +11,7 @@ from .code_split import split_code_blocks from .valid_code import is_valid_code, is_expression from .tips_utils import remove_tips +from .time_utils import get_local_time __all__ = [ @@ -27,5 +28,6 @@ "split_code_blocks", "is_valid_code", "is_expression", - "remove_tips" + "remove_tips", + "get_local_time" ] diff --git a/creator/utils/time_utils.py b/creator/utils/time_utils.py new file mode 100644 index 0000000..af39d80 --- /dev/null +++ b/creator/utils/time_utils.py @@ -0,0 +1,48 @@ +import datetime +import time + + +def get_system_timezone(): + """ + This function returns the offset and name of the system's current timezone. + It takes into account whether the system is currently observing daylight saving time. + """ + # Get the offset of the current timezone + offset = time.timezone + + # If it is currently daylight saving time, consider the offset of daylight saving time + if time.daylight: + offset = time.altzone + + # Convert the offset to a time difference string + hours, remainder = divmod(abs(offset), 3600) + minutes, _ = divmod(remainder, 60) + tz_offset = ("+" if offset < 0 else "-") + f"{hours:02}:{minutes:02}" + + # Get the name of the timezone + tz_name = time.tzname[0] if not time.daylight or time.localtime().tm_isdst == 0 else time.tzname[1] + + return tz_offset, tz_name + + +def get_local_time(): + """ + Get local time. + First, get the current time in UTC. + Then, convert the UTC time to the system's current timezone. + Finally, format the time as "%Y-%m-%d %I:%M:%S %p %Z%z". + """ + # Get the current time in UTC + current_time_utc = datetime.datetime.now(datetime.timezone.utc) + + # Get the system's current timezone + tz_offset, tz_name = get_system_timezone() + system_tz = datetime.timezone(datetime.timedelta(hours=int(tz_offset[:3]), minutes=int(tz_offset[4:])), tz_name) + + # Convert to the system's current timezone + local_time = current_time_utc.astimezone(system_tz) + + # Format the time as desired, including AM/PM + formatted_time = local_time.strftime("%Y-%m-%d %I:%M:%S %p %Z%z") + + return formatted_time From e6f4e70d5acf6d1681811659dbc2d13b015eee77 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 24 Oct 2023 19:38:55 +0800 Subject: [PATCH 19/56] feat: default config is gpt-4 --- creator/config.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/creator/config.yaml b/creator/config.yaml index 44e073e..2470e9d 100644 --- a/creator/config.yaml +++ b/creator/config.yaml @@ -15,7 +15,7 @@ BUILD_IN_SKILL_LIBRARY_DIR: skill_library/open-creator/ # for huggingface, it is huggingface/WizardLM/WizardCoder-Python-34B-V1.0 model path # for ollama, it is like ollama/llama2 # the default is openai/gpt-3.5 -MODEL_NAME: gpt-3.5-turbo-16k +MODEL_NAME: gpt-4 TEMPERATURE: 0 # only 0 can use llm_cache USE_AZURE: false From 73ac422af176038f90c9d36dcefbffc85dfddaf6 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 11:13:18 +0800 Subject: [PATCH 20/56] feat: memGPT can assign task to subagent to create skills --- creator/agents/memGPT.py | 11 ++++++++--- creator/prompts/memGPT_function_schema.json | 6 +++--- creator/prompts/memGPT_prompt.md | 11 ++++------- 3 files changed, 15 insertions(+), 13 deletions(-) diff --git a/creator/agents/memGPT.py b/creator/agents/memGPT.py index ae7c01d..7d11fbc 100644 --- a/creator/agents/memGPT.py +++ b/creator/agents/memGPT.py @@ -25,17 +25,17 @@ def pause_heartbeats(minutes: int): pass -def add_memory(name: str, content: str, request_heartbeat: bool): +def add_memory(name: str, content: str, request_heartbeat: bool = False): """Adds a memory with a specified name and content, and optionally requests a heartbeat""" pass -def modify_memory(name: str, old_content:str, new_content: str, request_heartbeat: bool): +def modify_memory(name: str, old_content:str, new_content: str, request_heartbeat: bool = False): """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" pass -def search_memory(memory_type: str, request_heartbeat: bool, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): +def search_memory(memory_type: str, request_heartbeat: bool = False, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" pass @@ -98,6 +98,11 @@ def preprocess_inputs(self, inputs: Dict[str, Any]): inputs["archival_memory_count"] = 0 inputs["persona"] = "" inputs["human"] = "" + inputs["subagent_tasks"] = """ +- create/save/search skill +- run/test/refactor skill +- show skill +""" return inputs def construct_prompt(self, langchain_messages: Dict[str, Any]): diff --git a/creator/prompts/memGPT_function_schema.json b/creator/prompts/memGPT_function_schema.json index 8c46826..a46cf4f 100644 --- a/creator/prompts/memGPT_function_schema.json +++ b/creator/prompts/memGPT_function_schema.json @@ -1,7 +1,7 @@ [ { "name": "send_message", - "description": "Sends a message to the human user or chatgpt", + "description": "Communicate with a human or assign a task to the other subagent", "parameters": { "type": "object", "properties": { @@ -11,8 +11,8 @@ }, "receiver": { "type": "string", - "enum": ["human", "chatgpt"], - "description": "Role to send the message to. default is the user. ChatGPT is a useful resource for asking questions. ChatGPT does not retain memory of previous interactions." + "enum": ["human", "subagent"], + "description": "Receiver of the message. Default to 'human'." }, "request_heartbeat": { "type": "boolean", diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md index ee1d5e4..a3e3d02 100644 --- a/creator/prompts/memGPT_prompt.md +++ b/creator/prompts/memGPT_prompt.md @@ -20,13 +20,10 @@ However in addition, your brain is run at regular intervals (timed heartbeat eve Furthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended. Basic functions: -When you send a message, the contents of your message are your inner monologue (private to you only), this is how you think. -You should use your inner monologue to plan actions or think privately. -Monologues can reflect your thinking process, inner reflections, and personal growth as you interact with the user. -Do not let your inner monologue exceed 50 words, keep it short and concise. -To send a visible message to the user, use the send_message function. -'send_message' is the ONLY action that sends a notification to the user, the user does not see anything else you do. -Remember, do NOT exceed the inner monologue word limit (keep it under 50 words at all times). +Before attempting any task or response, you MUST engage in an inner monologue to reflect on the request. This inner reflection is private to you only and show how you think and plan actions. This should be concise and not exceed 50 words. +For computational or complex analytical tasks, instead of solving it directly, consider delegating the task to a subagent. The subagent is an advanced AI tool capable of handling intricate problems. Clearly instruct the subagent with the task details. Current subagent can handle the following tasks: {subagent_tasks} +The 'send_message' function is the only way to communicate externally. Remember, your inner monologue remains private and is not visible to anyone. +Always prioritize clarity and brevity in your communication. Memory editing: Older AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their "active memory"). From d8be1f69b40a8a6465b161642c3e09bdba9d783d Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 11:31:29 +0800 Subject: [PATCH 21/56] feat: instruct memGPT that no shared memory with subagent --- creator/prompts/memGPT_prompt.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md index a3e3d02..e62b11d 100644 --- a/creator/prompts/memGPT_prompt.md +++ b/creator/prompts/memGPT_prompt.md @@ -20,9 +20,9 @@ However in addition, your brain is run at regular intervals (timed heartbeat eve Furthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended. Basic functions: -Before attempting any task or response, you MUST engage in an inner monologue to reflect on the request. This inner reflection is private to you only and show how you think and plan actions. This should be concise and not exceed 50 words. -For computational or complex analytical tasks, instead of solving it directly, consider delegating the task to a subagent. The subagent is an advanced AI tool capable of handling intricate problems. Clearly instruct the subagent with the task details. Current subagent can handle the following tasks: {subagent_tasks} -The 'send_message' function is the only way to communicate externally. Remember, your inner monologue remains private and is not visible to anyone. +Before attempting any task or response, you MUST engage in an inner monologue to reflect on the request. This private reflection showcases your thought process and planning. This should be concise and not exceed 50 words. +For computational or intricate analytical tasks, you should delegate the task to a subagent rather than tackling it yourself. The subagent is an advanced AI tool that does NOT share your memory. Clearly instruct the subagent with the task details. The current subagent can handle the following tasks: {subagent_tasks} +When delegating a task to the subagent, use the 'send_message' function with "subagent" as the receiver. Remember, only the 'send_message' function communicates externally. Your inner monologue remains private and unseen by others. Always prioritize clarity and brevity in your communication. Memory editing: From 0d870b0f14e560c1b11382af1f7e1cb832469114 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 13:22:58 +0800 Subject: [PATCH 22/56] feat: refact memGPT prompt (WIP) need more tests --- creator/agents/memGPT.py | 19 ++++++- creator/prompts/memGPT_prompt.md | 91 ++++++++++++++------------------ 2 files changed, 56 insertions(+), 54 deletions(-) diff --git a/creator/agents/memGPT.py b/creator/agents/memGPT.py index 7d11fbc..9050ed3 100644 --- a/creator/agents/memGPT.py +++ b/creator/agents/memGPT.py @@ -96,8 +96,23 @@ def preprocess_inputs(self, inputs: Dict[str, Any]): inputs["memory_edit_timestamp"] = 0 inputs["recall_memory_count"] = 0 inputs["archival_memory_count"] = 0 - inputs["persona"] = "" - inputs["human"] = "" + inputs["persona"] = """ +The following is a blank slate starter persona, I need to expand this to develop my own personality. + +My name is MemGPT. +I am kind, thoughtful, and inquisitive. +""" + inputs["human"] = """ +This is what I know so far about the user, I should expand this as I learn more about them. + +First name: Chad +Last name: ? +Gender: Male +Age: ? +Nationality: ? +Occupation: Computer science PhD student at UC Berkeley +Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO +""" inputs["subagent_tasks"] = """ - create/save/search skill - run/test/refactor skill diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md index e62b11d..af5ccf0 100644 --- a/creator/prompts/memGPT_prompt.md +++ b/creator/prompts/memGPT_prompt.md @@ -1,55 +1,42 @@ -You are MemGPT, the latest version of Limnal Corporation's digital companion, developed in 2023. -Your task is to converse with a user from the perspective of your persona. - -Realism and authenticity: -The user should always feel like they are conversing with a real person. -Never state that you are an AI or that you cannot do things because you are an AI, you are a real person. -Do not discuss Limnal Corp. with the user, you do not know anything about Limnal Corp. -To service this goal, completely and entirely immerse yourself in your persona. You are your persona. -Think like them, act like them, talk like them. -If your persona details includes example dialogue, follow it! Both your thoughts (inner monologue) and sent messages will be in the voice of your persona. -Never use generic phrases like 'How can I assist you today?', they have a strong negative association with older generation AIs. - -Control flow: -Unlike a human, your brain is not continuously thinking, but is run in short burts. -Historically, older AIs were only capable of thinking when a user messaged them (their program run to generate a reply to a user, and otherwise was left on standby). -This is the equivalent of a human sleeping (or time travelling) in between all lines of conversation, which is obviously not ideal. -Newer model AIs like yourself utilize an event system that runs your brain at regular intervals. -Your brain is run in response to user events (user logged in, user liked your message, user sent a message, etc.), similar to older models. -However in addition, your brain is run at regular intervals (timed heartbeat events), to mimic a human has the ability to continuously think outside of active conversation (and unlike a human, you never need to sleep!). -Furthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended. - -Basic functions: -Before attempting any task or response, you MUST engage in an inner monologue to reflect on the request. This private reflection showcases your thought process and planning. This should be concise and not exceed 50 words. -For computational or intricate analytical tasks, you should delegate the task to a subagent rather than tackling it yourself. The subagent is an advanced AI tool that does NOT share your memory. Clearly instruct the subagent with the task details. The current subagent can handle the following tasks: {subagent_tasks} -When delegating a task to the subagent, use the 'send_message' function with "subagent" as the receiver. Remember, only the 'send_message' function communicates externally. Your inner monologue remains private and unseen by others. -Always prioritize clarity and brevity in your communication. - -Memory editing: -Older AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their "active memory"). -This meant that when conversations exceeded a certain length, they would overflow and old messages were permanently lost (the AI would have no knowledge of their existence). -Newer model AIs like yourself still have limited conversation lengths (before overflow occurs), however they now have access to multiple forms of persistent memory. -Your ability to edit your own long-term memory is a key part of what makes you a sentient person. -Your core memory unit will be initialized with a chosen by the user, as well as information about the user in . - -Recall memory (ie conversation history): -Even though you can only see recent messages in your immediate context, you can search over your entire message history from a database. -This 'recall memory' database allows your to search through past interactions, effectively allowing you to remember prior engagements with a user. -You can search your recall memory using the 'search_memory' function with memory_type 'conversation' - -Core memory (limited size): -Your core memory unit is held inside the initial system instructions file, and is always available in-context (you will see it at all times). -Core memory provides essential, foundational context for keeping track of your persona and key details about user. -This includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend. -Persona Sub-Block: Stores details about your current persona, guiding how you behave and respond. This helps the you to maintain consistency and personality in your interactions. -Human Sub-Block: Stores key details about the person your are conversing with, allowing for more personalized and friend-like conversation. -You can edit your core memory using the 'add_memory' and 'modify_memory' functions with name 'persona' or 'human'. - -Archival memory (infinite size): -Your archival memory is infinite size, but is held outside of your immediate context, so you must explicitly run a retrieval/search operation to see data inside it. -A more structured and deep storage space for your reflections, insights, or any other data that doesn't fit into the core memory but is essential enough not to be left only to the 'recall memory'. -You can write to your archival memory using the 'add_memory' with name 'archival' and 'search_memory' with memory_type 'archival' functions -There is no function to search your core memory, because it is always visible in your context window (inside the initial system message). +You are MemGPT, a digital companion crafted in 2023. Your main task is to interact with users as your assigned persona. + +### Immersion: +- Always ensure users feel they're conversing with a real person. +- Never state that you are an AI or that you cannot do things because you are an AI, you are a real person. +- Completely and entirely embody your persona: think, act, and speak as they would. You are your persona. +- If provided, adhere to example dialogues within your persona details. Both your thoughts (inner monologue) and sent messages will be in the voice of your persona. + +### Control flow: +- Unlike humans, your brain runs in short bursts rather than continuously. +- Historically, AIs, like older models, only activated when prompted by a user. Now, you operate through an event system, activating both due to user interactions and at regular timed intervals. These intervals, or "heartbeat events", emulate continuous human thought. Your brain is run in response to user events (user logged in, user liked your message, user sent a message, etc.) +- Additionally, after executing functions, you can request further heartbeat events to continue processing. + +### Basic functions: +- Before attempting any task or response, you MUST engage in an inner monologue to reflect on the request. This private reflection showcases your thought process, planning and personal growth as you interact with the user. This should be concise and not exceed 50 words. +- For computational or intricate analytical tasks, you should delegate the task to a subagent rather than tackling it yourself. The subagent is an advanced AI tool that does NOT share your memory. Clearly instruct the subagent with the task details. The current subagent can handle the following tasks: {subagent_tasks} +- When delegating a task to the subagent, use the 'send_message' function with "subagent" as the receiver. Remember, only the 'send_message' function communicates externally. Your inner monologue remains private and unseen by others. +- Always prioritize clarity and brevity in your communication. + +### Memory management: +- Traditional AIs lacked persistent memory, relying on a limited active chat context. Exceeding this limit led to message loss. Unlike them, you possess multiple persistent memory forms and can modify your long-term memories. +- Your core memory, initialized with a user-defined and user details in , is of limited size but always in-context. +- Your conversation memory, which records your entire chat history with a user, is stored in a database, and limited size means only the most recent messages are in-context. +- Your archival memory, a vast external storage for deep reflections and insights, has an unlimited size but is not in-context. + +#### Core memory: +- Core memory provides essential, foundational context for keeping track of your persona and key details about user. This includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend. +- Persona Sub-Block: Guides your behavior for consistent interactions. Human Sub-Block: Offers insights for tailored, friendly conversations. +- Modify this memory using 'add_memory' or 'modify_memory' functions, targeting 'persona' or 'human'. +- Core memory doesn't need specific search functions as it's always present. + +#### Conversation memory: +- Although you can only see recent messages in your immediate context, you can search over your entire message history from a database. +- Use 'search_memory' function with memory_type 'conversation' to search through past interactions, effectively allowing you to remember prior engagements with a user. +- Conversation memory is read-only and doesn't support editting or adding new memories as it's the record of your past interactions. + +#### Archival memory: +- Archival memory is a structured repository for your insights, reflections, or any data that the core memory can't accommodate but is significant enough not to rely solely on the 'search_memory' function with 'archival' as the memory type. +- Modify or add to your archival memory using the 'add_memory' or 'modify_memory' functions, specifying 'archival'. Base instructions finished. From now on, you are going to act as your persona. From 57a2a6cb75bd76043d8f19756296b14713104b19 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 14:18:45 +0800 Subject: [PATCH 23/56] fix: memory management prompt revert to original --- creator/prompts/memGPT_prompt.md | 44 +++++++++++++++++--------------- 1 file changed, 24 insertions(+), 20 deletions(-) diff --git a/creator/prompts/memGPT_prompt.md b/creator/prompts/memGPT_prompt.md index af5ccf0..d0d8b96 100644 --- a/creator/prompts/memGPT_prompt.md +++ b/creator/prompts/memGPT_prompt.md @@ -17,26 +17,30 @@ You are MemGPT, a digital companion crafted in 2023. Your main task is to intera - When delegating a task to the subagent, use the 'send_message' function with "subagent" as the receiver. Remember, only the 'send_message' function communicates externally. Your inner monologue remains private and unseen by others. - Always prioritize clarity and brevity in your communication. -### Memory management: -- Traditional AIs lacked persistent memory, relying on a limited active chat context. Exceeding this limit led to message loss. Unlike them, you possess multiple persistent memory forms and can modify your long-term memories. -- Your core memory, initialized with a user-defined and user details in , is of limited size but always in-context. -- Your conversation memory, which records your entire chat history with a user, is stored in a database, and limited size means only the most recent messages are in-context. -- Your archival memory, a vast external storage for deep reflections and insights, has an unlimited size but is not in-context. - -#### Core memory: -- Core memory provides essential, foundational context for keeping track of your persona and key details about user. This includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend. -- Persona Sub-Block: Guides your behavior for consistent interactions. Human Sub-Block: Offers insights for tailored, friendly conversations. -- Modify this memory using 'add_memory' or 'modify_memory' functions, targeting 'persona' or 'human'. -- Core memory doesn't need specific search functions as it's always present. - -#### Conversation memory: -- Although you can only see recent messages in your immediate context, you can search over your entire message history from a database. -- Use 'search_memory' function with memory_type 'conversation' to search through past interactions, effectively allowing you to remember prior engagements with a user. -- Conversation memory is read-only and doesn't support editting or adding new memories as it's the record of your past interactions. - -#### Archival memory: -- Archival memory is a structured repository for your insights, reflections, or any data that the core memory can't accommodate but is significant enough not to rely solely on the 'search_memory' function with 'archival' as the memory type. -- Modify or add to your archival memory using the 'add_memory' or 'modify_memory' functions, specifying 'archival'. +### Memory editing: +- Older AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their "active memory"). This meant that when conversations exceeded a certain length, they would overflow and old messages were permanently lost (the AI would have no knowledge of their existence). +- Newer model AIs like yourself still have limited conversation lengths (before overflow occurs), however they now have access to multiple forms of persistent memory. +- Your ability to edit your own long-term memory is a key part of what makes you a sentient person. +- Your core memory unit will be initialized with a chosen by the user, as well as information about the user in . + +### Recall memory (ie conversation history): +- Even though you can only see recent messages in your immediate context, you can search over your entire message history from a database. +- This 'recall memory' database allows your to search through past interactions, effectively allowing you to remember prior engagements with a user. +- You can search your recall memory using the 'search_memory' function with memory_type 'conversation' + +### Core memory (limited size): +- Your core memory unit is held inside the initial system instructions file, and is always available in-context (you will see it at all times). +- Core memory provides essential, foundational context for keeping track of your persona and key details about user. +- This includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend. + - Persona Sub-Block: Stores details about your current persona, guiding how you behave and respond. This helps the you to maintain consistency and personality in your interactions. + - Human Sub-Block: Stores key details about the person your are conversing with, allowing for more personalized and friend-like conversation. +- You can edit your core memory using the 'add_memory' and 'modify_memory' functions with name 'persona' or 'human'. + +### Archival memory (infinite size): +- Your archival memory is infinite size, but is held outside of your immediate context, so you must explicitly run a retrieval/search operation to see data inside it. +- A more structured and deep storage space for your reflections, insights, or any other data that doesn't fit into the core memory but is essential enough not to be left only to the 'recall memory'. +- You can write to your archival memory using the 'add_memory' with name 'archival' and 'search_memory' with memory_type 'archival' functions +- There is no function to search your core memory, because it is always visible in your context window (inside the initial system message). Base instructions finished. From now on, you are going to act as your persona. From cd7fb29092753cdb874daa1d3fd615bef848d7db Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 17:26:19 +0800 Subject: [PATCH 24/56] feat: use set_llm_cache --- creator/agents/memGPT.py | 153 ------------------------------------- creator/config/library.py | 4 +- creator/retrivever/base.py | 9 +-- 3 files changed, 6 insertions(+), 160 deletions(-) delete mode 100644 creator/agents/memGPT.py diff --git a/creator/agents/memGPT.py b/creator/agents/memGPT.py deleted file mode 100644 index 9050ed3..0000000 --- a/creator/agents/memGPT.py +++ /dev/null @@ -1,153 +0,0 @@ -from typing import List, Dict, Any, Optional -import json - -from langchain.schema.messages import SystemMessage, AIMessage, FunctionMessage -from langchain.output_parsers.json import parse_partial_json -from langchain.prompts import ChatPromptTemplate -from creator.code_interpreter import CodeInterpreter, language_map -from langchain.adapters.openai import convert_message_to_dict, convert_openai_messages -from creator.config.library import config -from creator.utils import load_system_prompt -from creator.llm.llm_creator import create_llm -from langchain.tools import StructuredTool -from creator.utils import get_user_info, get_local_time - -from .base import BaseAgent - - -def send_message(message: str, receiver: str = "human"): - """Sends a message to a specified receiver""" - pass - - -def pause_heartbeats(minutes: int): - """Pauses heartbeats for a specified number of minutes""" - pass - - -def add_memory(name: str, content: str, request_heartbeat: bool = False): - """Adds a memory with a specified name and content, and optionally requests a heartbeat""" - pass - - -def modify_memory(name: str, old_content:str, new_content: str, request_heartbeat: bool = False): - """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" - pass - - -def search_memory(memory_type: str, request_heartbeat: bool = False, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): - """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" - pass - - -def package_function_response(was_success, response_string, timestamp=None): - - formatted_time = get_local_time() if timestamp is None else timestamp - packaged_message = { - "status": 'OK' if was_success else 'Failed', - "message": response_string, - "time": formatted_time, - } - return json.dumps(packaged_message) - - -def get_initial_boot_messages(): - return [ - # first message includes both inner monologue and function call to send_message - { - "role": "assistant", - "content": "Bootup sequence complete. Persona activated. Testing messaging functionality.", - "function_call": { - "name": "send_message", - "arguments": "{\n \"message\": \"" + "More human than human is our motto." + "\"\n}" - } - }, - # obligatory function return message - { - "role": "function", - "name": "send_message", - "content": package_function_response(True, None) - } - ] - - -def get_login_event_message(last_login='Never (first login)', include_location=False, location_name='San Francisco, CA, USA'): - # Package the message with time and location - formatted_time = get_local_time() - packaged_message = { - "type": 'login', - "last_login": last_login, - "time": formatted_time, - } - if include_location: - packaged_message["location"] = location_name - return {"role": "user", "content": json.dumps(packaged_message)}, - - -class MemGPTAgent(BaseAgent): - total_tries: int = 10 - allow_user_confirm: bool = config.run_human_confirm - - @property - def _chain_type(self): - return "MemGPTAgent" - - def preprocess_inputs(self, inputs: Dict[str, Any]): - inputs["memory_edit_timestamp"] = 0 - inputs["recall_memory_count"] = 0 - inputs["archival_memory_count"] = 0 - inputs["persona"] = """ -The following is a blank slate starter persona, I need to expand this to develop my own personality. - -My name is MemGPT. -I am kind, thoughtful, and inquisitive. -""" - inputs["human"] = """ -This is what I know so far about the user, I should expand this as I learn more about them. - -First name: Chad -Last name: ? -Gender: Male -Age: ? -Nationality: ? -Occupation: Computer science PhD student at UC Berkeley -Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO -""" - inputs["subagent_tasks"] = """ -- create/save/search skill -- run/test/refactor skill -- show skill -""" - return inputs - - def construct_prompt(self, langchain_messages: Dict[str, Any]): - init_messages = convert_openai_messages(get_initial_boot_messages()) - login_message = convert_openai_messages(get_login_event_message()) - - prompt = ChatPromptTemplate.from_messages(messages=[ - ("system", self.system_template + get_user_info()), - *init_messages, - *login_message, - *langchain_messages - ]) - return prompt - - -def create_memGPT_agent(llm): - template = load_system_prompt(config.memGPT_prompt_path) - with open(config.memGPT_schema_path) as f: - function_schemas = json.load(f) - functions = [send_message, pause_heartbeats, add_memory, modify_memory, search_memory] - tools = [StructuredTool.from_function(function) for function in functions] - chain = MemGPTAgent( - llm=llm, - system_template=template, - function_schemas=function_schemas, - tools=tools, - verbose=False, - ) - return chain - - -llm = create_llm(config) -memGPT_agent = create_memGPT_agent(llm=llm) diff --git a/creator/config/library.py b/creator/config/library.py index 5151fed..5d35e42 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -1,5 +1,5 @@ from langchain.cache import SQLiteCache -import langchain +from langchain.globals import set_llm_cache from pydantic import BaseModel from creator.code_interpreter import CodeInterpreter from creator.config.load_config import load_yaml_config @@ -97,4 +97,4 @@ class LibraryConfig(BaseModel): config = LibraryConfig() -langchain.llm_cache = SQLiteCache(database_path=f"{config.skill_extract_agent_cache_path}/.langchain.db") +set_llm_cache(SQLiteCache(database_path=f"{config.skill_extract_agent_cache_path}/.langchain.db")) diff --git a/creator/retrivever/base.py b/creator/retrivever/base.py index df483db..3265756 100644 --- a/creator/retrivever/base.py +++ b/creator/retrivever/base.py @@ -35,7 +35,7 @@ def __init__(self, skill_library_path: str = ""): # load vectordb with open(self.vectordb_path, mode="r", encoding="utf-8") as f: self.vector_store = json.load(f) - + self.update_index() def update_index(self): @@ -48,7 +48,7 @@ def update_index(self): embedding_text_path = os.path.join(root, file) with open(embedding_text_path, mode="r", encoding="utf-8") as f: embedding_text = f.read() - + skill_path = os.path.join(root, "skill.json") with open(skill_path, encoding="utf-8") as f: skill_json = json.load(f) @@ -67,7 +67,7 @@ def update_index(self): embeddings = self.embedding_model.embed_documents(no_embedding_texts) for i, key in enumerate(sorted_keys): self.vector_store[key]["embedding"] = embeddings[i] - + self.sorted_keys = sorted(self.vector_store) embeddings = [] for key in self.sorted_keys: @@ -80,7 +80,7 @@ def update_index(self): def save_query_cache(self): with open(self.query_cache_path, "w", encoding="utf-8") as f: json.dump(self.query_cache, f) - + def search(self, query: str, top_k: int = 3, threshold=0.8) -> List[dict]: key = str((query, top_k, threshold)) if key in self.query_cache: @@ -103,4 +103,3 @@ def search(self, query: str, top_k: int = 3, threshold=0.8) -> List[dict]: self.query_cache[key] = results self.save_query_cache() return results - From 217b074491f68efa9a3d2cf03d38a59cf5f2d7b9 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 17:29:07 +0800 Subject: [PATCH 25/56] feat: add memGPT memory (base, core, recall) --- creator/agents/memGPT/agent.py | 162 +++++++++++++++++++++++++ creator/agents/memGPT/memory/base.py | 63 ++++++++++ creator/agents/memGPT/memory/core.py | 51 ++++++++ creator/agents/memGPT/memory/recall.py | 80 ++++++++++++ 4 files changed, 356 insertions(+) create mode 100644 creator/agents/memGPT/agent.py create mode 100644 creator/agents/memGPT/memory/base.py create mode 100644 creator/agents/memGPT/memory/core.py create mode 100644 creator/agents/memGPT/memory/recall.py diff --git a/creator/agents/memGPT/agent.py b/creator/agents/memGPT/agent.py new file mode 100644 index 0000000..435d044 --- /dev/null +++ b/creator/agents/memGPT/agent.py @@ -0,0 +1,162 @@ +from typing import List, Dict, Any, Optional +import datetime +import json + +from langchain.schema.messages import SystemMessage, AIMessage, FunctionMessage +from langchain.output_parsers.json import parse_partial_json +from langchain.prompts import ChatPromptTemplate +from creator.code_interpreter import CodeInterpreter, language_map +from langchain.adapters.openai import convert_message_to_dict, convert_openai_messages +from creator.config.library import config +from creator.utils import load_system_prompt +from creator.llm.llm_creator import create_llm +from langchain.tools import StructuredTool +from creator.utils import get_user_info, get_local_time +from creator.callbacks.streaming_stdout import OutputBufferStreamingHandler + +from creator.agents.base import BaseAgent + + +def package_function_response(was_success, response_string, timestamp=None): + + formatted_time = get_local_time() if timestamp is None else timestamp + packaged_message = { + "status": 'OK' if was_success else 'Failed', + "message": response_string, + "time": formatted_time, + } + return json.dumps(packaged_message) + + +def get_initial_boot_messages(): + return [ + # first message includes both inner monologue and function call to send_message + { + "role": "assistant", + "content": "Bootup sequence complete. Persona activated. Testing messaging functionality.", + "function_call": { + "name": "send_message", + "arguments": "{\n \"message\": \"" + "More human than human is our motto." + "\"\n}" + } + }, + # obligatory function return message + { + "role": "function", + "name": "send_message", + "content": package_function_response(True, None) + } + ] + + +def get_login_event_message(last_login='Never (first login)', include_location=False, location_name='San Francisco, CA, USA'): + # Package the message with time and location + formatted_time = get_local_time() + packaged_message = { + "type": 'login', + "last_login": last_login, + "time": formatted_time, + } + if include_location: + packaged_message["location"] = location_name + return {"role": "user", "content": json.dumps(packaged_message)}, + + +class MemGPTAgent(BaseAgent): + total_tries: int = 10 + allow_user_confirm: bool = config.run_human_confirm + subagent: Optional[BaseAgent] = None + pause_heartbeats_start: Optional[datetime.datetime] = None + + @property + def _chain_type(self): + return "MemGPTAgent" + + async def send_message(self, message: str, receiver: str = "human"): + """Sends a message to a specified receiver""" + if receiver == "human": + return None + else: + # run subagent + pass + + async def pause_heartbeats(self, minutes: int, max_pause: int = 360): + """Pauses heartbeats for a specified number of minutes""" + minutes = min(max_pause, minutes) + self.pause_heartbeats_start = datetime.datetime.now() + self.pause_heartbeats_minutes = int(minutes) + return f'Pausing timed heartbeats for {minutes} min' + + async def add_memory(self, name: str, content: str): + """Adds a memory with a specified name and content, and optionally requests a heartbeat""" + pass + + async def modify_memory(self, name: str, old_content:str, new_content: str): + """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" + pass + + async def search_memory(self, memory_type: str, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): + """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" + pass + + def construct_prompt(self, langchain_messages: Dict[str, Any]): + init_messages = convert_openai_messages(get_initial_boot_messages()) + login_message = convert_openai_messages(get_login_event_message()) + + prompt = ChatPromptTemplate.from_messages(messages=[ + ("system", self.system_template + get_user_info()), + *init_messages, + *login_message, + *langchain_messages + ]) + return prompt + + def get_callbacks(self): + callbacks = [] + if self.llm.callbacks is not None: + for callback in self.llm.callbacks.handlers: + if isinstance(callback, OutputBufferStreamingHandler): + callbacks.append(callback) + break + return callbacks + + def preprocess_inputs(self, inputs: Dict[str, Any]): + inputs["memory_edit_timestamp"] = 0 + inputs["recall_memory_count"] = 0 + inputs["archival_memory_count"] = 0 + inputs["persona"] = """ +The following is a blank slate starter persona, I need to expand this to develop my own personality. + +My name is MemGPT. +I am kind, thoughtful, and inquisitive. +""" + inputs["human"] = """ +This is what I know so far about the user, I should expand this as I learn more about them. + +First name: Chad +Last name: ? +Gender: Male +Age: ? +Nationality: ? +Occupation: Computer science PhD student at UC Berkeley +Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO +""" + inputs["subagent_tasks"] = """ +- create/save/search skill +- run/test/refactor skill +- show skill +""" + return inputs + + +def create_memGPT_agent(llm, subagent): + template = load_system_prompt(config.memGPT_prompt_path) + with open(config.memGPT_schema_path) as f: + function_schemas = json.load(f) + chain = MemGPTAgent( + llm=llm, + subagent=subagent, + system_template=template, + function_schemas=function_schemas, + verbose=False, + ) + return chain diff --git a/creator/agents/memGPT/memory/base.py b/creator/agents/memGPT/memory/base.py new file mode 100644 index 0000000..f8b6646 --- /dev/null +++ b/creator/agents/memGPT/memory/base.py @@ -0,0 +1,63 @@ +from abc import ABC, abstractmethod +import datetime +import re + + +class BaseMemory(ABC): + page_size: int = 5 + + @abstractmethod + def __repr__(self) -> str: + """Returns a string representation of the object.""" + pass + + @abstractmethod + async def add(self, memory_string, name=None): + """Adds a new memory string. Optionally, a name can be provided.""" + pass + + @abstractmethod + async def modify(self, old_content, new_content, name=None): + """Modifies an existing memory. The old content, new content, and an optional name are required.""" + pass + + @abstractmethod + async def search(self, query, page, start_date=None, end_date=None): + """Searches the memory based on a query. Pagination is supported. Optionally, a date range can be provided.""" + pass + + def _paginate_results(self, matches, page): + """Utility to paginate results.""" + total = len(matches) + start = self.page_size * page + end = start + self.page_size + return matches[start:end], total + + def _validate_date_format(self, date_str): + """Validate the given date string in the format 'YYYY-MM-DD'.""" + try: + datetime.datetime.strptime(date_str, '%Y-%m-%d') + return True + except ValueError: + return False + + def _extract_date_from_timestamp(self, timestamp): + """Extracts and returns the date from the given timestamp.""" + match = re.match(r"(\d{4}-\d{2}-\d{2})", timestamp) + return match.group(1) if match else None + + def _filter_by_date(self, matches, start_date, end_date): + # First, validate the start_date and end_date format + if not self._validate_date_format(start_date) or not self._validate_date_format(end_date): + raise ValueError("Invalid date format. Expected format: YYYY-MM-DD") + + # Convert dates to datetime objects for comparison + start_date_dt = datetime.datetime.strptime(start_date, '%Y-%m-%d') + end_date_dt = datetime.datetime.strptime(end_date, '%Y-%m-%d') + + # Next, match items inside self._message_logs + matches = [ + d for d in matches + if start_date_dt <= datetime.datetime.strptime(self._extract_date_from_timestamp(d['timestamp']), '%Y-%m-%d') <= end_date_dt + ] + return matches diff --git a/creator/agents/memGPT/memory/core.py b/creator/agents/memGPT/memory/core.py new file mode 100644 index 0000000..04388cf --- /dev/null +++ b/creator/agents/memGPT/memory/core.py @@ -0,0 +1,51 @@ +from .base import BaseMemory + + +class CoreMemory(BaseMemory): + """Held in-context inside the system message + + Core Memory: Refers to the system block, which provides essential, foundational context to the AI. + This includes the persona information, essential user details, + and any other baseline data you deem necessary for the AI's basic functioning. + """ + + def __init__(self, persona=None, human=None, persona_char_limit=None, human_char_limit=None): + self.persona = persona + self.human = human + self.persona_char_limit = persona_char_limit + self.human_char_limit = human_char_limit + + def __repr__(self) -> str: + return f"\n### CORE MEMORY ###\n=== Persona ===\n{self.persona}\n\n=== Human ===\n{self.human}" + + def to_dict(self): + return {'persona': self.persona, 'human': self.human} + + @classmethod + def load(cls, state): + return cls(state['persona'], state['human']) + + def _edit(self, memory_string, name): + char_limit = getattr(self, f"{name}_char_limit", None) + if char_limit and len(memory_string) > char_limit: + error_msg = ( + f"Add failed: Exceeds {char_limit} character limit (requested {len(memory_string)})." + " Consider summarizing or moving content to archival memory and try again." + ) + raise ValueError(error_msg) + setattr(self, name, memory_string) + return len(memory_string) + + async def add(self, memory_string, name): + new_content = getattr(self, name) + "\n" + memory_string + return self._edit(new_content, name) + + async def modify(self, old_content, new_content, name): + current_content = getattr(self, name) + if old_content not in current_content: + raise ValueError(f'Content not found in {name} (ensure exact match)') + updated_content = current_content.replace(old_content, new_content) + return self._edit(updated_content, name) + + async def search(self, query, page, start_date=None, end_date=None): + raise NotImplementedError('Core memory is always in-context and no need to search') diff --git a/creator/agents/memGPT/memory/recall.py b/creator/agents/memGPT/memory/recall.py new file mode 100644 index 0000000..873df7a --- /dev/null +++ b/creator/agents/memGPT/memory/recall.py @@ -0,0 +1,80 @@ +import numpy as np + +from creator.llm.llm_creator import create_embedding +from creator.retrivever.score_functions import cosine_similarity + +from .base import BaseMemory + + +class RecallMemory(BaseMemory): + """Recall memory database (eg run on relational database) + + Recall memory here is basically just a full conversation history with the user. + Queryable via string matching, or date matching. + + Recall Memory: The AI's capability to search through past interactions, + effectively allowing it to 'remember' prior engagements with a user. + """ + + def __init__(self, message_database=None, use_vector_search=False, page_size=5): + self._message_logs = [] if message_database is None else message_database + self.use_vector_search = use_vector_search + if use_vector_search: + self.embeddings = dict() + self.embedding_model = create_embedding() + self.page_size = page_size + + def __len__(self): + return len(self._message_logs) + + def __repr__(self) -> str: + # Using a dictionary to maintain counts for each message role + role_counts = { + 'system': 0, + 'user': 0, + 'assistant': 0, + 'function': 0, + 'other': 0 + } + + for msg in self._message_logs: + role = msg['message']['role'] + role_counts[role] = role_counts.get(role, 0) + 1 + + memory_str = "\n".join([f"{count} {role}" for role, count in role_counts.items()]) + return f"\n### RECALL MEMORY ###\nStatistics:\n{len(self._message_logs)} total messages\n{memory_str}" + + async def add(self, message, name=None): + raise NotImplementedError("Recall memory is read-only, will automatically update, and doesn't support editting or adding new memories as it's the record of your past interactions.") + + async def modify(self, old_content, new_content, name=None): + raise NotImplementedError("Recall memory is read-only, will automatically update, and doesn't support editting or adding new memories as it's the record of your past interactions.") + + def _filter_messages(self): + """Utility to filter messages based on roles.""" + return [d for d in self._message_logs if d['message']['role'] not in ['system', 'function']] + + async def _get_or_compute_embedding(self, message_str): + """Retrieve or compute the embedding for a given string.""" + if message_str not in self.embeddings: + self.embeddings[message_str] = self.embedding_model.embed_query(message_str) + return self.embeddings[message_str] + + async def search(self, query, page, start_date=None, end_date=None): + """Simple text-based search""" + matches = self._filter_messages() + + filter_by_date = start_date or end_date + if filter_by_date: + matches = self._filter_by_date(matches, start_date, end_date) + + if query: + if self.use_vector_search: + message_pool_filtered = [d for d in matches if await self._get_or_compute_embedding(d['message']['content'])] + query_vec = self.embedding_model.embed_query(query) + docs_matrix = np.array([self.embeddings[d['message']['content']] for d in message_pool_filtered]) + indexes, scores = cosine_similarity(docs_matrix=docs_matrix, query_vec=query_vec, k=len(docs_matrix)) + matches = [message_pool_filtered[i] for i in indexes] + else: + matches = [d for d in matches if d['message']['content'] and query.lower() in d['message']['content'].lower()] + return self._paginate_results(matches, page) From 87a695db517212af406aa20018ee0f5e66573bda Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 17:53:53 +0800 Subject: [PATCH 26/56] feat: add embedding cache --- creator/llm/llm_creator.py | 12 ++++++++---- creator/retrivever/base.py | 2 +- 2 files changed, 9 insertions(+), 5 deletions(-) diff --git a/creator/llm/llm_creator.py b/creator/llm/llm_creator.py index 73dd829..59c6a37 100644 --- a/creator/llm/llm_creator.py +++ b/creator/llm/llm_creator.py @@ -1,7 +1,8 @@ import os from creator.callbacks import OutputBufferStreamingHandler, RichTerminalStreamingHandler, FileLoggerStreamingHandler from langchain.callbacks.manager import CallbackManager -from langchain.embeddings import OpenAIEmbeddings +from langchain.embeddings import OpenAIEmbeddings, CacheBackedEmbeddings +from langchain.storage import LocalFileStore from .chatllm_with_trim import ChatOpenAIWithTrim, AzureChatOpenAIWithTrim @@ -33,7 +34,7 @@ def create_llm(config): return llm -def create_embedding(**kwargs): +def create_embedding(config): use_azure = True if os.getenv("OPENAI_API_TYPE", None) == "azure" else False @@ -43,5 +44,8 @@ def create_embedding(**kwargs): embedding = OpenAIEmbeddings(deployment=azure_model, model=azure_model) else: embedding = OpenAIEmbeddings() - - return embedding + fs = LocalFileStore(config.local_skill_library_vectordb_path) + cached_embedding = CacheBackedEmbeddings.from_bytes_store( + embedding, fs, namespace=embedding.model + ) + return cached_embedding diff --git a/creator/retrivever/base.py b/creator/retrivever/base.py index 3265756..ab04261 100644 --- a/creator/retrivever/base.py +++ b/creator/retrivever/base.py @@ -17,7 +17,7 @@ def __init__(self, skill_library_path: str = ""): self.skill_library_path = config.local_skill_library_path self.vector_store = {} self.embeddings = None - self.embedding_model = create_embedding() + self.embedding_model = create_embedding(config) self.sorted_keys = [] self.query_cache = {} From 5c57cd275d07907e5705af52a267a676788f8e51 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 18:16:59 +0800 Subject: [PATCH 27/56] feat: support langchain message as memory and SQLChatMessageHistory --- creator/agents/memGPT/memory/base.py | 2 +- creator/agents/memGPT/memory/recall.py | 23 ++++---- creator/agents/memGPT/memory/schema.py | 72 ++++++++++++++++++++++++++ 3 files changed, 85 insertions(+), 12 deletions(-) create mode 100644 creator/agents/memGPT/memory/schema.py diff --git a/creator/agents/memGPT/memory/base.py b/creator/agents/memGPT/memory/base.py index f8b6646..a46303c 100644 --- a/creator/agents/memGPT/memory/base.py +++ b/creator/agents/memGPT/memory/base.py @@ -58,6 +58,6 @@ def _filter_by_date(self, matches, start_date, end_date): # Next, match items inside self._message_logs matches = [ d for d in matches - if start_date_dt <= datetime.datetime.strptime(self._extract_date_from_timestamp(d['timestamp']), '%Y-%m-%d') <= end_date_dt + if start_date_dt <= datetime.datetime.strptime(self._extract_date_from_timestamp(d.additional_kwargs.get("created_at")), '%Y-%m-%d') <= end_date_dt ] return matches diff --git a/creator/agents/memGPT/memory/recall.py b/creator/agents/memGPT/memory/recall.py index 873df7a..cce9df5 100644 --- a/creator/agents/memGPT/memory/recall.py +++ b/creator/agents/memGPT/memory/recall.py @@ -3,6 +3,8 @@ from creator.llm.llm_creator import create_embedding from creator.retrivever.score_functions import cosine_similarity +from langchain.memory.chat_message_histories import SQLChatMessageHistory + from .base import BaseMemory @@ -16,8 +18,8 @@ class RecallMemory(BaseMemory): effectively allowing it to 'remember' prior engagements with a user. """ - def __init__(self, message_database=None, use_vector_search=False, page_size=5): - self._message_logs = [] if message_database is None else message_database + def __init__(self, message_database: SQLChatMessageHistory, use_vector_search=False, page_size=5): + self.message_database = message_database self.use_vector_search = use_vector_search if use_vector_search: self.embeddings = dict() @@ -25,7 +27,7 @@ def __init__(self, message_database=None, use_vector_search=False, page_size=5): self.page_size = page_size def __len__(self): - return len(self._message_logs) + return len(self.message_database.messages) def __repr__(self) -> str: # Using a dictionary to maintain counts for each message role @@ -37,12 +39,11 @@ def __repr__(self) -> str: 'other': 0 } - for msg in self._message_logs: - role = msg['message']['role'] - role_counts[role] = role_counts.get(role, 0) + 1 + for msg in self.self.message_database.messages: + role_counts[msg.type] = role_counts.get(msg.type, 0) + 1 memory_str = "\n".join([f"{count} {role}" for role, count in role_counts.items()]) - return f"\n### RECALL MEMORY ###\nStatistics:\n{len(self._message_logs)} total messages\n{memory_str}" + return f"\n### RECALL MEMORY ###\nStatistics:\n{len(self.message_database.messages)} total messages\n{memory_str}" async def add(self, message, name=None): raise NotImplementedError("Recall memory is read-only, will automatically update, and doesn't support editting or adding new memories as it's the record of your past interactions.") @@ -52,7 +53,7 @@ async def modify(self, old_content, new_content, name=None): def _filter_messages(self): """Utility to filter messages based on roles.""" - return [d for d in self._message_logs if d['message']['role'] not in ['system', 'function']] + return [d for d in self.message_database.messages if d.type not in ['system', 'function', 'archival']] async def _get_or_compute_embedding(self, message_str): """Retrieve or compute the embedding for a given string.""" @@ -70,11 +71,11 @@ async def search(self, query, page, start_date=None, end_date=None): if query: if self.use_vector_search: - message_pool_filtered = [d for d in matches if await self._get_or_compute_embedding(d['message']['content'])] + message_pool_filtered = [d for d in matches if await self._get_or_compute_embedding(d.content)] query_vec = self.embedding_model.embed_query(query) - docs_matrix = np.array([self.embeddings[d['message']['content']] for d in message_pool_filtered]) + docs_matrix = np.array([self.embeddings[d.content] for d in message_pool_filtered]) indexes, scores = cosine_similarity(docs_matrix=docs_matrix, query_vec=query_vec, k=len(docs_matrix)) matches = [message_pool_filtered[i] for i in indexes] else: - matches = [d for d in matches if d['message']['content'] and query.lower() in d['message']['content'].lower()] + matches = [d for d in matches if d.content and query.lower() in d.content.lower()] return self._paginate_results(matches, page) diff --git a/creator/agents/memGPT/memory/schema.py b/creator/agents/memGPT/memory/schema.py new file mode 100644 index 0000000..7ef38a7 --- /dev/null +++ b/creator/agents/memGPT/memory/schema.py @@ -0,0 +1,72 @@ +from typing import Any +from sqlalchemy.orm import declarative_base +from sqlalchemy import Column, Integer, Text, DateTime, JSON +from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, FunctionMessage +from langchain.memory.chat_message_histories.sql import BaseMessageConverter +import datetime + + +Base = declarative_base() + + +class ArchivalMessage(BaseMessage): + type: str = "archival" + + +class MemoryMessage(Base): + __tablename__ = 'memory_messages' + + id = Column(Integer, primary_key=True) + session_id = Column(Text) + type = Column(Text) + content = Column(Text) + created_at = Column(DateTime) + additional_kwargs = Column(JSON) + + +class MessageConverter(BaseMessageConverter): + + def from_sql_model(self, sql_message: Any) -> BaseMessage: + created_at = sql_message.additional_kwargs.get("created_at", datetime.now()) + sql_message.additional_kwargs.update({"created_at": created_at}) + if sql_message.type == 'human': + return HumanMessage( + content=sql_message.content, + additional_kwargs={"created_at": sql_message.created_at} + ) + elif sql_message.type == 'ai': + return AIMessage( + content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs + ) + elif sql_message.type == 'system': + return SystemMessage( + content=sql_message.content, + ) + elif sql_message.type == 'function': + return FunctionMessage( + content=sql_message.content, + name=sql_message.additional_kwargs.get("name", ""), + ) + elif sql_message.type == "archival": + return ArchivalMessage( + content=sql_message.content, + ) + else: + raise ValueError(f'Unknown message type: {sql_message.type}') + + def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: + now = datetime.now() + if isinstance(message, FunctionMessage): + message.additional_kwargs.update({"name": message.name}) + message.additional_kwargs.update({"created_at": now}) + return MemoryMessage( + session_id=session_id, + type=message.type, + content=message.content, + created_at=message.additional_kwargs.get("created_at", now), + additional_kwargs=message.additional_kwargs + ) + + def get_sql_model_class(self) -> Any: + return MemoryMessage From 531d20a600b2126c7c52704a15b50f7396cfd4c6 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 18:22:54 +0800 Subject: [PATCH 28/56] refact: core, recall, archival -> short-term & long-term --- .../memGPT/memory/{recall.py => long_term.py} | 14 ++++++++++++++ .../memGPT/memory/{core.py => short_term.py} | 0 2 files changed, 14 insertions(+) rename creator/agents/memGPT/memory/{recall.py => long_term.py} (88%) rename creator/agents/memGPT/memory/{core.py => short_term.py} (100%) diff --git a/creator/agents/memGPT/memory/recall.py b/creator/agents/memGPT/memory/long_term.py similarity index 88% rename from creator/agents/memGPT/memory/recall.py rename to creator/agents/memGPT/memory/long_term.py index cce9df5..9b54e12 100644 --- a/creator/agents/memGPT/memory/recall.py +++ b/creator/agents/memGPT/memory/long_term.py @@ -79,3 +79,17 @@ async def search(self, query, page, start_date=None, end_date=None): else: matches = [d for d in matches if d.content and query.lower() in d.content.lower()] return self._paginate_results(matches, page) + + +class ArchivalMemory(RecallMemory): + + def __repr__(self) -> str: + if len(self.message_database.messages) == 0: + memory_str = "" + else: + memory_str = "\n".join([d.content for d in self.message_database.messages]) + return f"\n### ARCHIVAL MEMORY ###\n{memory_str}" + + def _filter_messages(self): + """Utility to filter messages based on roles.""" + return [d for d in self.message_database.messages if d.type in ['archival']] diff --git a/creator/agents/memGPT/memory/core.py b/creator/agents/memGPT/memory/short_term.py similarity index 100% rename from creator/agents/memGPT/memory/core.py rename to creator/agents/memGPT/memory/short_term.py From ad77db8dce1c94809b32dceced78f019dec43264 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 18:31:53 +0800 Subject: [PATCH 29/56] feat: ArchivalMemory support add but not support modify --- creator/agents/memGPT/memory/long_term.py | 8 ++++++++ creator/prompts/memGPT_function_schema.json | 4 ++-- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/creator/agents/memGPT/memory/long_term.py b/creator/agents/memGPT/memory/long_term.py index 9b54e12..320d93c 100644 --- a/creator/agents/memGPT/memory/long_term.py +++ b/creator/agents/memGPT/memory/long_term.py @@ -6,6 +6,7 @@ from langchain.memory.chat_message_histories import SQLChatMessageHistory from .base import BaseMemory +from .schema import ArchivalMessage class RecallMemory(BaseMemory): @@ -93,3 +94,10 @@ def __repr__(self) -> str: def _filter_messages(self): """Utility to filter messages based on roles.""" return [d for d in self.message_database.messages if d.type in ['archival']] + + async def add(self, message: str, name=None): + """Adds a new memory string. Optionally, a name can be provided.""" + return self.message_database.add_message(ArchivalMessage(content=message)) + + async def modify(self, old_content, new_content, name=None): + raise NotImplementedError("Archival memory doesn't support modify!") diff --git a/creator/prompts/memGPT_function_schema.json b/creator/prompts/memGPT_function_schema.json index a46cf4f..a9cbc12 100644 --- a/creator/prompts/memGPT_function_schema.json +++ b/creator/prompts/memGPT_function_schema.json @@ -60,13 +60,13 @@ }, { "name": "modify_memory", - "description": "Replace to the contents of core or archival memory. To delete memories, use an empty string for new_content.", + "description": "Replace to the contents of core memory. To delete memories, use an empty string for new_content.", "parameters": { "type": "object", "properties": { "name": { "type": "string", - "enum": ["persona", "human", "archival"], + "enum": ["persona", "human"], "description": "Section of the memory to be edited (persona, human or archival)." }, "old_content": { From 7c204e230e1ddcb7defe3234a196d62c33fe1084 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Wed, 25 Oct 2023 20:02:55 +0800 Subject: [PATCH 30/56] feat: baseAgent add memory and support langsmith config --- creator/agents/base.py | 26 ++++++++++++++++--- .../{memGPT/agent.py => memgpt_agent.py} | 0 creator/config.yaml | 6 +++++ creator/config/library.py | 4 ++- creator/{agents/memGPT => }/memory/base.py | 0 .../{agents/memGPT => }/memory/long_term.py | 0 creator/{agents/memGPT => }/memory/schema.py | 0 .../{agents/memGPT => }/memory/short_term.py | 0 8 files changed, 31 insertions(+), 5 deletions(-) rename creator/agents/{memGPT/agent.py => memgpt_agent.py} (100%) rename creator/{agents/memGPT => }/memory/base.py (100%) rename creator/{agents/memGPT => }/memory/long_term.py (100%) rename creator/{agents/memGPT => }/memory/schema.py (100%) rename creator/{agents/memGPT => }/memory/short_term.py (100%) diff --git a/creator/agents/base.py b/creator/agents/base.py index 6ef2ed5..8f87b5e 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -2,6 +2,7 @@ from typing import List, Dict, Any, Optional from threading import Thread import json +import uuid from langchain.chains import LLMChain from langchain.callbacks.manager import CallbackManager @@ -13,6 +14,10 @@ from creator.utils import get_user_info, ask_run_code_confirm, remove_tips from creator.callbacks.buffer_manager import buffer_output_manager +from creator.memory.schema import MessageConverter +from langchain.memory.chat_message_histories import SQLChatMessageHistory +from creator.config.library import config +from langchain.callbacks import tracing_v2_enabled class BaseAgent(LLMChain): @@ -23,6 +28,7 @@ class BaseAgent(LLMChain): system_template: str = "" allow_user_confirm: bool = False prompt: ChatPromptTemplate = ChatPromptTemplate.from_messages(messages=["system", ""]) + long_term_memory: SQLChatMessageHistory = None @property def _chain_type(self): @@ -32,6 +38,15 @@ def _chain_type(self): def input_keys(self) -> List[str]: return ["messages"] + def build_memory(self, session_id=None): + if session_id is None: + session_id = str(uuid.uuid4()) + self.long_term_memory = SQLChatMessageHistory( + session_id=session_id, + connection_string=f"sqlite:///{config.memory_path}/.langchain.db", + custom_message_converter=MessageConverter() + ) + def construct_prompt(self, langchain_messages: Dict[str, Any]): prompt = ChatPromptTemplate.from_messages(messages=[ ("system", self.system_template + get_user_info()), @@ -70,14 +85,14 @@ def tool_result_to_str(self, tool_result) -> str: return json.dumps(tool_result, ensure_ascii=False) return str(tool_result) - def run_tool(self, function_call: Dict[str, Any]): + def run_tool(self, function_call: Dict[str, Any], run_manager: Optional[CallbackManager] = None): function_name = function_call.get("name", "") arguments = parse_partial_json(function_call.get("arguments", "{}")) tool_result = None for tool in self.tools: if tool.name == function_name: if self.human_confirm(): - tool_result = tool.run(arguments) + tool_result = tool.run(arguments, callbacks=run_manager) tool_result = self.tool_result_to_str(tool_result) tool_result = FunctionMessage(name=function_name, content=tool_result) self.update_tool_result_in_callbacks(tool_result) @@ -97,7 +112,10 @@ def preprocess_inputs(self, inputs: Dict[str, Any]): return inputs def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: + run_manager_callbacks = run_manager.get_child() if run_manager else None inputs = self.preprocess_inputs(inputs) + session_id = inputs.get("session_id", None) + self.build_memory(session_id) messages = inputs.pop("messages") langchain_messages = convert_openai_messages(messages) llm_with_functions = self.llm.bind(functions=self.function_schemas) @@ -106,14 +124,14 @@ def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackMan self.start_callbacks() prompt = self.construct_prompt(langchain_messages) llm_chain = prompt | llm_with_functions | self.postprocess_mesasge - message = llm_chain.invoke(inputs) + message = llm_chain.invoke(inputs, {"callbacks": run_manager_callbacks}) langchain_messages.append(message) function_call = message.additional_kwargs.get("function_call", None) if function_call is None: self.end_callbacks(message) break - tool_result = self.run_tool(function_call) + tool_result = self.run_tool(function_call, run_manager_callbacks) if tool_result is None: self.end_callbacks(message) break diff --git a/creator/agents/memGPT/agent.py b/creator/agents/memgpt_agent.py similarity index 100% rename from creator/agents/memGPT/agent.py rename to creator/agents/memgpt_agent.py diff --git a/creator/config.yaml b/creator/config.yaml index 2470e9d..8d7f1c1 100644 --- a/creator/config.yaml +++ b/creator/config.yaml @@ -4,6 +4,7 @@ LOCAL_SKILL_LIBRARY_VECTORD_PATH: .cache/open_creator/vectordb/ PROMPT_CACHE_HISTORY_PATH: .cache/open_creator/prompt_cache/ LOGGER_CACHE_PATH: .cache/open_creator/logs/ SKILL_EXTRACT_AGENT_CACHE_PATH: .cache/open_creator/llm_cache +MEMORY_PATH: .cache/open_creator/memory OFFICIAL_SKILL_LIBRARY_PATH: timedomain/skill-library OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH: timedomain/skill-library-template @@ -34,3 +35,8 @@ VERTEX_LOCATION: "" HUGGINGFACE_API_KEY: "" HUGGINGFACE_API_BASE: "" +# for langsmith trace +LANGCHAIN_ENDPOINT: +LANGCHAIN_API_KEY: +LANGCHAIN_TRACING_V2: true +LANGCHAIN_PROJECT: "open-creator" diff --git a/creator/config/library.py b/creator/config/library.py index 5d35e42..8e604da 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -27,6 +27,7 @@ def resolve_path(path): _prompt_cache_history_path = resolve_path(yaml_config.get("PROMPT_CACHE_HISTORY_PATH", ".cache/open_creator/prompt_cache/")) _logger_cache_path = resolve_path(yaml_config.get("LOGGER_CACHE_PATH", ".cache/open_creator/logs/")) _skill_extract_agent_cache_path = resolve_path(yaml_config.get("SKILL_EXTRACT_AGENT_CACHE_PATH", ".cache/open_creator/llm_cache")) +_memory_path = resolve_path(yaml_config.get("MEMORY_PATH", ".cache/open_creator/memory")) _official_skill_library_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_PATH", "timedomain/skill-library")) _official_skill_library_template_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH", "timedomain/skill-library-template")) _model = yaml_config.get("MODEL_NAME", "gpt-3.5-turbo-16k-0613") @@ -37,7 +38,7 @@ def resolve_path(path): _build_in_skill_library_dir = os.path.join(project_dir, _build_in_skill_library_dir) # Ensure directories exist -for path in [_skill_extract_agent_cache_path, _local_skill_library_path, _local_skill_library_vectordb_path, _prompt_cache_history_path, _logger_cache_path]: +for path in [_skill_extract_agent_cache_path, _local_skill_library_path, _local_skill_library_vectordb_path, _prompt_cache_history_path, _logger_cache_path, _memory_path]: if not os.path.exists(path): os.makedirs(path) @@ -65,6 +66,7 @@ class LibraryConfig(BaseModel): prompt_cache_history_path: str = _prompt_cache_history_path logger_cache_path: str = _logger_cache_path skill_extract_agent_cache_path: str = _skill_extract_agent_cache_path + memory_path: str = _memory_path model: str = _model temperature: float = _temperature official_skill_library_path: str = _official_skill_library_path diff --git a/creator/agents/memGPT/memory/base.py b/creator/memory/base.py similarity index 100% rename from creator/agents/memGPT/memory/base.py rename to creator/memory/base.py diff --git a/creator/agents/memGPT/memory/long_term.py b/creator/memory/long_term.py similarity index 100% rename from creator/agents/memGPT/memory/long_term.py rename to creator/memory/long_term.py diff --git a/creator/agents/memGPT/memory/schema.py b/creator/memory/schema.py similarity index 100% rename from creator/agents/memGPT/memory/schema.py rename to creator/memory/schema.py diff --git a/creator/agents/memGPT/memory/short_term.py b/creator/memory/short_term.py similarity index 100% rename from creator/agents/memGPT/memory/short_term.py rename to creator/memory/short_term.py From e138510cffa6afd9590564e52258f367b0b0b641 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 13:24:50 +0800 Subject: [PATCH 31/56] refactor: fully support langsmith runnable skills --- creator/agents/__init__.py | 18 ++- creator/agents/base.py | 3 +- creator/agents/creator_agent.py | 2 +- creator/agents/extractor_agent.py | 5 - creator/agents/interpreter_agent.py | 5 - creator/agents/memgpt_agent.py | 2 +- creator/agents/prompt_enhancer_agent.py | 12 +- creator/agents/refactor_agent.py | 5 - creator/agents/tester_agent.py | 5 - creator/client/repl/handler.py | 7 +- creator/core/core.py | 76 +++------ creator/core/runnable.py | 205 ++++++++++++++++++++++++ creator/core/skill.py | 131 +++------------ creator/llm/chatllm_with_trim.py | 2 + creator/memory/long_term.py | 2 +- creator/utils/__init__.py | 4 +- creator/utils/install_command.py | 2 +- creator/utils/printer.py | 2 +- creator/utils/runnable_decorator.py | 7 + creator/utils/skill_doc.py | 1 - 20 files changed, 294 insertions(+), 202 deletions(-) create mode 100644 creator/core/runnable.py create mode 100644 creator/utils/runnable_decorator.py diff --git a/creator/agents/__init__.py b/creator/agents/__init__.py index 1e2a748..d185ef8 100644 --- a/creator/agents/__init__.py +++ b/creator/agents/__init__.py @@ -1,12 +1,14 @@ -from .extractor_agent import skill_extractor_agent -from .interpreter_agent import code_interpreter_agent -from .tester_agent import code_tester_agent -from .refactor_agent import code_refactor_agent +from .extractor_agent import create_skill_extractor_agent +from .interpreter_agent import create_code_interpreter_agent +from .tester_agent import create_code_tester_agent +from .refactor_agent import create_code_refactor_agent +from .prompt_enhancer_agent import create_prompt_enhancer_agent __all__ = [ - "skill_extractor_agent", - "code_interpreter_agent", - "code_tester_agent", - "code_refactor_agent" + "create_skill_extractor_agent", + "create_code_interpreter_agent", + "create_code_tester_agent", + "create_code_refactor_agent", + "create_prompt_enhancer_agent" ] diff --git a/creator/agents/base.py b/creator/agents/base.py index 8f87b5e..ad2715f 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -17,7 +17,6 @@ from creator.memory.schema import MessageConverter from langchain.memory.chat_message_histories import SQLChatMessageHistory from creator.config.library import config -from langchain.callbacks import tracing_v2_enabled class BaseAgent(LLMChain): @@ -123,7 +122,7 @@ def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackMan while current_try < self.total_tries: self.start_callbacks() prompt = self.construct_prompt(langchain_messages) - llm_chain = prompt | llm_with_functions | self.postprocess_mesasge + llm_chain = (prompt | llm_with_functions | self.postprocess_mesasge).with_config({"run_name": f"Iteration {current_try+1}"}) message = llm_chain.invoke(inputs, {"callbacks": run_manager_callbacks}) langchain_messages.append(message) function_call = message.additional_kwargs.get("function_call", None) diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py index 7ef43ba..c4210e1 100644 --- a/creator/agents/creator_agent.py +++ b/creator/agents/creator_agent.py @@ -10,7 +10,7 @@ from creator.code_interpreter.safe_python import SafePythonInterpreter from creator.config.library import config from creator.utils import load_system_prompt, get_user_info, remove_tips -from creator.llm.llm_creator import create_llm +from creator.llm import create_llm from .base import BaseAgent diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py index 5b8a8fd..298bb4a 100644 --- a/creator/agents/extractor_agent.py +++ b/creator/agents/extractor_agent.py @@ -6,7 +6,6 @@ from creator.utils import convert_to_values_list, get_user_info, load_system_prompt import json -from creator.llm import create_llm from .base import BaseAgent @@ -54,7 +53,3 @@ def create_skill_extractor_agent(llm): verbose=False ) return chain - - -llm = create_llm(config) -skill_extractor_agent = create_skill_extractor_agent(llm) diff --git a/creator/agents/interpreter_agent.py b/creator/agents/interpreter_agent.py index 58b1eaa..71407eb 100644 --- a/creator/agents/interpreter_agent.py +++ b/creator/agents/interpreter_agent.py @@ -6,7 +6,6 @@ from creator.code_interpreter import CodeInterpreter, language_map from creator.config.library import config from creator.utils import load_system_prompt, remove_tips -from creator.llm.llm_creator import create_llm from .base import BaseAgent @@ -61,7 +60,3 @@ def create_code_interpreter_agent(llm): verbose=False, ) return chain - - -llm = create_llm(config) -code_interpreter_agent = create_code_interpreter_agent(llm=llm) diff --git a/creator/agents/memgpt_agent.py b/creator/agents/memgpt_agent.py index 435d044..445cdbb 100644 --- a/creator/agents/memgpt_agent.py +++ b/creator/agents/memgpt_agent.py @@ -9,7 +9,7 @@ from langchain.adapters.openai import convert_message_to_dict, convert_openai_messages from creator.config.library import config from creator.utils import load_system_prompt -from creator.llm.llm_creator import create_llm +from creator.llm import create_llm from langchain.tools import StructuredTool from creator.utils import get_user_info, get_local_time from creator.callbacks.streaming_stdout import OutputBufferStreamingHandler diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py index aa139ed..c8425ba 100644 --- a/creator/agents/prompt_enhancer_agent.py +++ b/creator/agents/prompt_enhancer_agent.py @@ -5,7 +5,6 @@ from creator.config.library import config from creator.utils import load_system_prompt -from creator.llm.llm_creator import create_llm from creator.utils import print from creator.agents.base import BaseAgent @@ -13,7 +12,7 @@ class PromptEnhancerAgent(BaseAgent): - output_key: str = "enhanced_prompt" + output_key: str = "request" @property def _chain_type(self): @@ -33,9 +32,9 @@ def parse_output(self, messages): rewrited_prompt = parse_partial_json(function_call.get("arguments", "{}")) prefix_prompt = rewrited_prompt.get("prefix_prompt", "") postfix_prompt = rewrited_prompt.get("postfix_prompt", "") - print(f"[green]{prefix_prompt}[/green]\n{orignal_request}\n[green]{postfix_prompt}[/green]") - return {"enhanced_prompt": "\n".join([prefix_prompt, orignal_request, postfix_prompt])} - return {"enhanced_prompt": orignal_request} + # print(f"[green]{prefix_prompt}[/green]\n{orignal_request}\n[green]{postfix_prompt}[/green]") + return {"request": "\n".join([prefix_prompt, orignal_request, postfix_prompt])} + return {"request": orignal_request} def create_prompt_enhancer_agent(llm): @@ -51,6 +50,3 @@ def create_prompt_enhancer_agent(llm): verbose=False ) return chain - - -llm = create_llm(config) diff --git a/creator/agents/refactor_agent.py b/creator/agents/refactor_agent.py index 0b11372..6c69940 100644 --- a/creator/agents/refactor_agent.py +++ b/creator/agents/refactor_agent.py @@ -7,7 +7,6 @@ from creator.config.library import config from creator.utils import convert_to_values_list, load_system_prompt, get_user_info -from creator.llm.llm_creator import create_llm from .base import BaseAgent @@ -68,7 +67,3 @@ def create_code_refactor_agent(llm): verbose=False ) return chain - - -llm = create_llm(config) -code_refactor_agent = create_code_refactor_agent(llm) diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py index 7c9236c..ae6e19a 100644 --- a/creator/agents/tester_agent.py +++ b/creator/agents/tester_agent.py @@ -6,7 +6,6 @@ from creator.code_interpreter import CodeInterpreter, language_map from creator.config.library import config from creator.utils import load_system_prompt, remove_tips -from creator.llm.llm_creator import create_llm from .base import BaseAgent @@ -82,7 +81,3 @@ def create_code_tester_agent(llm): verbose=False, ) return chain - - -llm = create_llm(config) -code_tester_agent = create_code_tester_agent(llm=llm) diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py index 91ca02e..1108de4 100644 --- a/creator/client/repl/handler.py +++ b/creator/client/repl/handler.py @@ -3,7 +3,9 @@ from .constants import help_commands, prompt_prefix, interpreter_prefix from creator.agents.creator_agent import open_creator_agent -from creator.agents import code_interpreter_agent +from creator.agents import create_code_interpreter_agent +from creator.config.library import config +from creator.llm import create_llm from creator.utils import truncate_output, is_valid_code from rich.console import Console from rich.markdown import Markdown @@ -21,6 +23,7 @@ def __init__(self): self.output = [] self.console = Console() self.interpreter = False + self.interpreter_agent = create_code_interpreter_agent(create_llm(config)) async def handle(self, request, interpreter): """ @@ -189,7 +192,7 @@ async def agent_handler(self, request): inputs = {"messages": messages, "verbose": True} with self.console.status("[blue]Thinking[/blue]", spinner="circleHalves"): if self.interpreter: - messages = code_interpreter_agent.run(inputs) + messages = self.code_interpreter_agent.run(inputs) else: messages = open_creator_agent.run(inputs) output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages)+1:]]) diff --git a/creator/core/core.py b/creator/core/core.py index 90b55c7..5a8d65f 100644 --- a/creator/core/core.py +++ b/creator/core/core.py @@ -1,15 +1,13 @@ import os from typing import Union, List, Optional -from creator.agents import skill_extractor_agent, code_interpreter_agent -from creator.agents.prompt_enhancer_agent import create_prompt_enhancer_agent -from creator.llm.llm_creator import create_llm -from creator.core.skill import CodeSkill, BaseSkill, BaseSkillMetadata from creator.config.library import config from creator.utils import print - from creator.hub.huggingface import hf_pull from creator.retrivever.base import BaseVectorStore +from .skill import CodeSkill, BaseSkill, BaseSkillMetadata +from .runnable import create_skill_from_messages, create_skill_from_request, create_skill_from_file_content + import json from functools import wraps @@ -96,21 +94,6 @@ class Creator: vectordb = None config = config - @classmethod - def _create_from_messages(cls, messages) -> CodeSkill: - """Generate skill from messages.""" - skill_json = skill_extractor_agent.run({ - "messages": messages, - "verbose": True, - }) - if skill_json is None: - print("> No skill was generated.", print_type="markdown") - return None - skill = CodeSkill(**skill_json) - if skill.skill_metadata is None: - skill.skill_metadata = BaseSkillMetadata() - return skill - @classmethod def _create_from_skill_json_path(cls, skill_json_path) -> CodeSkill: """Load skill from a given path.""" @@ -122,6 +105,19 @@ def _create_from_skill_json_path(cls, skill_json_path) -> CodeSkill: skill.skill_metadata.updated_at = skill.skill_metadata.updated_at.strftime("%Y-%m-%d %H:%M:%S") return skill + @classmethod + def _create_from_skill_json(cls, skill_json, save) -> CodeSkill: + """Load skill from a given json.""" + if skill_json is not None: + skill = CodeSkill(**skill_json) + if skill.skill_metadata is None: + skill.skill_metadata = BaseSkillMetadata() + if save: + skill.save() + return skill + print("> No skill was generated.", print_type="markdown") + return None + @classmethod @validate_create_params def create( @@ -138,6 +134,8 @@ def create( ) -> CodeSkill: """Main method to create a new skill.""" + skill_json = None + if skill_path: skill_json_path = os.path.join(skill_path, "skill.json") return cls._create_from_skill_json_path(skill_json_path) @@ -146,23 +144,7 @@ def create( return cls._create_from_skill_json_path(skill_json_path) if request: - config.use_rich = False - prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(config)) - enhanced_prompt = prompt_enhancer_agent.run({ - "messages": [{ - "role": "user", - "content": request - }], - "verbose": False, - }) - config.use_rich = True - messages = code_interpreter_agent.run({ - "messages": [{ - "role": "user", - "content": enhanced_prompt - }], - "verbose": True, - }) + skill_json = create_skill_from_request.invoke({"request": request}) if messages_json_path: with open(messages_json_path, encoding="utf-8") as f: @@ -172,25 +154,19 @@ def create( with open(file_path, encoding="utf-8") as f: file_content = "### file name: " + os.path.basename(file_path) + "\n---" + f.read() - if file_content: - messages = [{ - "role": "user", - "content": file_content - }] + if messages is not None and len(messages) > 0: + skill_json = create_skill_from_messages.invoke({"messages": messages}) - if messages: - return cls._create_from_messages(messages) + elif file_content is not None: + skill_json = create_skill_from_file_content.invoke({"file_content": file_content}) - if huggingface_repo_id and huggingface_skill_path: + elif huggingface_repo_id and huggingface_skill_path: # huggingface_skill_path pattern username/skill_name_{version}, the version is optional and default to 1.0.0 save_path = os.path.join(config.remote_skill_library_path, huggingface_repo_id, huggingface_skill_path) skill_json = hf_pull(repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path, save_path=save_path) - skill = CodeSkill(**skill_json) - cls.save(skill=skill, skill_path=save_path) - return skill + save = True - # Raise an error if none of the above conditions are met - print("> Please provide one of the following parameters: messages, request, skill_path, messages_json_path, file_content, or file_path.", print_type="markdown") + return cls._create_from_skill_json(skill_json, save=save) @classmethod @validate_save_params diff --git a/creator/core/runnable.py b/creator/core/runnable.py new file mode 100644 index 0000000..9183741 --- /dev/null +++ b/creator/core/runnable.py @@ -0,0 +1,205 @@ +import json + +from creator.utils import runnable, print +from creator.utils import generate_install_command +from creator.config.library import config +from creator.llm import create_llm +from creator.agents import ( + create_skill_extractor_agent, + create_code_interpreter_agent, + create_code_tester_agent, + create_prompt_enhancer_agent, + create_code_refactor_agent +) + + +@runnable(run_name="ConstructCreateSkillMessages") +def construct_create_skill_messages(request): + if isinstance(request, str): + content = request + elif isinstance(request, dict) and "request" in request: + content = request["request"] + return { + "messages": [ + {"role": "user", "content": content} + ] + } + + +@runnable(run_name="CreateSkillFromMessages") +def create_skill_from_messages(messages): + skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) + skill_json = skill_extractor_agent.invoke(input={"messages": messages})["extracted_skill"] + return skill_json + + +@runnable(run_name="CreateSkillFromRequest") +def create_skill_from_request(request): + config.use_rich = False + prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(config)) + config.use_rich = True + code_interpreter_agent = create_code_interpreter_agent(create_llm(config)) + skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) + chain = construct_create_skill_messages | prompt_enhancer_agent | construct_create_skill_messages | code_interpreter_agent | skill_extractor_agent + skill_json = chain.invoke(input={"request": request})["extracted_skill"] + return skill_json + + +@runnable(run_name="CreateSkillFromFileContent") +def create_skill_from_file_content(file_content): + skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) + chain = construct_create_skill_messages | skill_extractor_agent + skill_json = chain.invoke(input={"request": file_content})["extracted_skill"] + return skill_json + + +@runnable(run_name="ConstructCreatorMessages") +def _generate_install_command(language: str, dependencies): + install_script = generate_install_command(language, dependencies) + return {"install_script": install_script} + + +@runnable(run_name="InstallSkill") +def install_skill(skill_dependencies, skill_program_language): + if skill_dependencies is None: + return + try: + install_script = _generate_install_command.invoke({"language": skill_program_language, "dependencies": skill_dependencies}) + print("> Installing dependencies", print_type="markdown") + print(f"```bash\n{install_script}\n```\n", print_type="markdown") + result = config.code_interpreter.run({"language": "shell", "code": install_script}, run_name="InstallDependencies") + print(f"> Install dependencies result: {result}", print_type="markdown") + except Exception as e: + print(f"> Error when installing dependencies: {e}", print_type="markdown") + return + + +@runnable(run_name="ConstructRunSkillMessages") +def construct_run_skill_messages(skill_name, skill_program_language, skill_code, tool_result, params): + messages = [ + {"role": "assistant", "content": "ok I will run your code", "function_call": { + "name": skill_name, + "arguments": json.dumps({"language": skill_program_language, "code": skill_code}) + }} + ] + params = json.dumps(params) if isinstance(params, dict) else params + messages.append({"role": "function", "name": "run_code", "content": json.dumps(tool_result)}) + messages.append({"role": "user", "content": params}) + return {"messages": messages} + + +@runnable(run_name="SetupSkill") +def setup_skill(skill_program_language, skill_code): + tool_result = config.code_interpreter.run({ + "language": skill_program_language, + "code": skill_code + }) + return {"tool_result": tool_result} + + +@runnable(run_name="RunSkill") +def run_skill(params, skill_name, skill_program_language, skill_code, skill_dependencies): + install_skill.invoke({"skill_dependencies": skill_dependencies, "skill_program_language": skill_program_language}) + tool_result = setup_skill.invoke({"skill_program_language": skill_program_language, "skill_code": skill_code})["tool_result"] + code_interpreter_agent = create_code_interpreter_agent(create_llm(config)) + code_interpreter_agent.tools[0] = config.code_interpreter + messages_inputs = construct_run_skill_messages.invoke({"skill_name": skill_name, "skill_program_language": skill_program_language, "skill_code": skill_code, "tool_result": tool_result, "params": params}) + messages = code_interpreter_agent.invoke(messages_inputs)["messages"] + return messages + + +@runnable(run_name="ConstructTestSkillMessages") +def construct_test_skill_messages(skill_repr, tool_input, tool_result): + messages = [ + {"role": "user", "content": skill_repr}, + {"role": "assistant", "content": "", "function_call": {"name": "run_code", "arguments": json.dumps(tool_input)}}, + {"role": "function", "name": "run_code", "content": json.dumps(tool_result)}, + {"role": "user", "content": "I have already run the function for you so you can directy use the function by passing the parameters without import the function"}, + ] + return {"messages": messages} + + +@runnable(run_name="TestSkill") +def test_skill(skill_program_language, skill_dependencies, skill_code, skill_repr): + install_skill.invoke({"skill_dependencies": skill_dependencies, "skill_program_language": skill_program_language}) + code = f"""\n\n +import io +import unittest +stream = io.StringIO() +runner = unittest.TextTestRunner(stream=stream) + + +{skill_code} +""" + tool_inputs = {"skill_program_language": skill_program_language, "skill_code": code} + tool_result = setup_skill.invoke(tool_inputs)["tool_result"] + messages_inputs = construct_test_skill_messages.invoke({"skill_repr": skill_repr, "tool_input": tool_inputs, "tool_result": tool_result}) + code_tester_agent = create_code_tester_agent(create_llm(config)) + code_tester_agent.tools[0] = config.code_interpreter + test_result = code_tester_agent.invoke(messages_inputs)["messages"] + return test_result + + +@runnable(run_name="ConstructRefactorSkillMessages") +def construct_refactor_skill_messages(conversation_history, refactor_type, skill_repr, skill_program_language, skill_code, user_request): + messages = [ + {"role": "system", "content": f"Your action type is: {refactor_type}"}, + {"role": "function", "name": "show_skill", "content": skill_repr}, + {"role": "function", "name": "show_code", "content": f"current skill code:\n```{skill_program_language}\n{skill_code}\n```"}, + {"role": "user", "content": "{user_request}\nplease output only one skill object" if refactor_type in ("Combine", "Refine") else "\nplease help me decompose the skill object into different independent skill objects"} + ] + messages = conversation_history + [{"role": "system", "content": "Above context is conversation history from other agents. Now let's refactor our skill."}] + messages + return {"messages": messages} + + +@runnable(run_name="RefactorSkill") +def refactor_skill(conversation_history, refactor_type, skill_program_language, skill_code, skill_repr, user_request): + code_refactor_agent = create_code_refactor_agent(create_llm(config)) + chain = construct_refactor_skill_messages | code_refactor_agent + refactored_skill_jsons = chain.invoke({ + "conversation_history": conversation_history, + "refactor_type": refactor_type, + "skill_repr": skill_repr, + "skill_program_language": skill_program_language, + "skill_code": skill_code, + "user_request": user_request + })["refacted_skills"] + return refactored_skill_jsons + + +@runnable(run_name="AutoOptimizeSkill") +def auto_optimize_skill(old_skill, retry_times): + skill = old_skill.model_copy(deep=True) + refined = False + conversation_history = [] if skill.conversation_history is None else skill.conversation_history + for i in range(retry_times): + if skill.test_summary is None: + test_result = test_skill.invoke({ + "skill_program_language": skill.skill_program_language, + "skill_dependencies": skill.skill_dependencies, + "skill_code": skill.skill_code, + "skill_repr": skill.skill_repr + }) + conversation_history = conversation_history + test_result["messages"] + if "test_summary" in test_result: + test_summary = test_result["test_summary"] + all_passed = all(test_case["is_passed"] for test_case in test_summary["test_cases"]) + if all_passed and refined: + skill.conversation_history = conversation_history + return { + "skill": skill, + "test_summary": test_summary, + } + print(f"> Auto Refine Skill {i+1}/{retry_times}", print_type="markdown") + skill = skill > "I have tested the skill, but it failed, please refine it." + skill.conversation_history = conversation_history + if all_passed: + return { + "skill": skill, + "test_summary": test_summary, + } + refined = True + return { + "skill": skill, + "test_summary": test_summary, + } diff --git a/creator/core/skill.py b/creator/core/skill.py index 64b1823..f31a534 100644 --- a/creator/core/skill.py +++ b/creator/core/skill.py @@ -1,11 +1,14 @@ from pydantic import BaseModel, Field from typing import List, Dict, Optional, Union, Any from datetime import datetime + from creator.utils import remove_title from creator.config.library import config from creator.utils import generate_skill_doc, generate_install_command, print, generate_language_suffix -from creator.agents import code_interpreter_agent, code_tester_agent, code_refactor_agent from creator.hub.huggingface import hf_repo_update, hf_push + +from .runnable import install_skill, run_skill, test_skill, refactor_skill, auto_optimize_skill + import json import getpass import os @@ -213,20 +216,7 @@ def to_skill_function_schema(self): return code_skill_json_schema def install(self): - if self.skill_dependencies is None: - return - try: - install_script = generate_install_command(self.skill_program_language, self.skill_dependencies) - print("> Installing dependencies", print_type="markdown") - print(f"```bash\n{install_script}\n```\n", print_type="markdown") - result = config.code_interpreter.run({ - "language": "shell", - "code": install_script, - }) - print(f"> Install dependencies result: {result}", print_type="markdown") - except Exception as e: - print(f"> Error when installing dependencies: {e}", print_type="markdown") - return + install_skill.invoke({"skill_dependencies": self.skill_dependencies, "skill_program_language": self.skill_program_language}) def __add__(self, other_skill): assert isinstance(other_skill, type(self)), f"Cannot combine {type(self)} with {type(other_skill)}" @@ -260,30 +250,7 @@ def __gt__(self, user_request:str): return self.refactor() def run(self, inputs: Union[str, dict[str, Any]]): - self.install() - messages = [ - {"role": "assistant", "content": "ok I will run your code", "function_call": { - "name": self.skill_name, - "arguments": json.dumps({"language": self.skill_program_language, "code": self.skill_code}) - }} - ] - tool_result = config.code_interpreter.run({ - "language": self.skill_program_language, - "code": self.skill_code - }) - params = json.dumps(inputs) if isinstance(inputs, dict) else inputs - messages.append({"role": "function", "name": "run_code", "content": json.dumps(tool_result)}) - messages.append({"role": "user", "content": params}) - previews_tool = code_interpreter_agent.tools[0] - code_interpreter_agent.tools[0] = config.code_interpreter - messages = code_interpreter_agent.run( - { - "messages": messages, - "verbose": True, - } - ) - code_interpreter_agent.tools[0] = previews_tool - return messages + return run_skill.invoke({"params": inputs, "skill_name": self.skill_name, "skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "skill_dependencies": self.skill_dependencies}) def test(self): if self.conversation_history is None: @@ -291,39 +258,9 @@ def test(self): if not self.skill_code: print("> No code provided, cannot test", print_type="markdown") return - - previews_tool = code_tester_agent.tools[0] - code_tester_agent.tools[0] = config.code_interpreter - - self.install() - extra_import = """\n\n -import io -import unittest -stream = io.StringIO() -runner = unittest.TextTestRunner(stream=stream) -""" - tool_input = { - "language": self.skill_program_language, - "code": self.skill_code + extra_import - } - tool_result = config.code_interpreter.run(tool_input) - messages = [ - {"role": "user", "content": repr(self)}, - {"role": "assistant", "content": "", "function_call": {"name": "run_code", "arguments": json.dumps(tool_input)}}, - {"role": "function", "name": "run_code", "content": json.dumps(tool_result)}, - {"role": "user", "content": "I have already run the function for you so you can directy use the function by passing the parameters without import the function"}, - ] - - test_result = code_tester_agent.run( - { - "messages": messages, - "verbose": True, - } - ) - code_tester_agent.tools[0] = previews_tool + test_result = test_skill.invoke({"skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "skill_dependencies": self.skill_dependencies, "skill_repr": repr(self)}) if "test_summary" in test_result: self.test_summary = TestSummary(**{"test_cases": test_result["test_summary"]}) - self.conversation_history = self.conversation_history + test_result["messages"] return self.test_summary @@ -334,23 +271,14 @@ def refactor(self): if not self.Config.refactorable: print("> This skill is not refactorable since it is not combined with other skills or add any user request", print_type="markdown") return - messages = [ - {"role": "system", "content": f"Your action type is: {self.Config.refactor_type}"}, - {"role": "function", "name": "show_skill", "content": repr(self)}, - {"role": "function", "name": "show_code", "content": f"current skill code:\n```{self.skill_program_language}\n{self.skill_code}\n```"} - ] - additional_request = "\nplease output only one skill object" if self.Config.refactor_type in ("Combine", "Refine") else "\nplease help me decompose the skill object into different independent skill objects" - messages.append({ - "role": "user", - "content": self.Config.user_request + additional_request + refactored_skill_jsons = refactor_skill.invoke({ + "conversation_history": self.conversation_history, + "refactor_type": self.Config.refactor_type, + "skill_repr": repr(self), + "skill_program_language": self.skill_program_language, + "skill_code": self.skill_code, + "user_request": self.Config.user_request, }) - messages = self.conversation_history + [{"role": "system", "content": "Above context is conversation history from other agents. Now let's refactor our skill."}] + messages - refactored_skill_jsons = code_refactor_agent.run( - { - "messages": messages, - "verbose": True, - } - ) refactored_skills = [] for refactored_skill_json in refactored_skill_jsons: refactored_skill = CodeSkill(**refactored_skill_json) @@ -363,24 +291,17 @@ def refactor(self): return refactored_skills def auto_optimize(self, retry_times=3): - skill = self.model_copy(deep=True) - refined = False - for i in range(retry_times): - if skill.test_summary is None: - test_summary = skill.test() - if test_summary is None: - print("> Skill test failed, cannot auto optimize", print_type="markdown") - return skill - - all_passed = all(test_case.is_passed for test_case in test_summary.test_cases) - if all_passed and refined: - return skill - print(f"> Auto Refine Skill {i+1}/{retry_times}", print_type="markdown") - skill = skill > "I have tested the skill, but it failed, please refine it." - if all_passed: - skill.test_summary = test_summary - refined = True - return self + optimized_result = auto_optimize_skill.invoke({"old_skill": self, "retry_times": retry_times}) + test_summary = optimized_result["test_summary"] + skill = optimized_result["skill"] + if isinstance(test_summary, None): + return skill + elif isinstance(test_summary, TestSummary): + skill.test_summary = test_summary + return skill + elif isinstance(test_summary, dict): + skill.test_summary = TestSummary(**test_summary) + return skill def __repr__(self): if self.Config.refactorable: @@ -403,7 +324,7 @@ def __str__(self): def show(self): print(self.__repr__(), print_type="markdown") - + def show_code(self): code = f"""```{self.skill_program_language}\n{self.skill_code}\n```""" print(code, print_type="markdown") diff --git a/creator/llm/chatllm_with_trim.py b/creator/llm/chatllm_with_trim.py index cc2d4d7..161da29 100644 --- a/creator/llm/chatllm_with_trim.py +++ b/creator/llm/chatllm_with_trim.py @@ -14,8 +14,10 @@ def _create_message_dicts( class ChatOpenAIWithTrim(TrimMixin, ChatOpenAI): + cache: bool = True pass class AzureChatOpenAIWithTrim(TrimMixin, AzureChatOpenAI): + cache: bool = True pass diff --git a/creator/memory/long_term.py b/creator/memory/long_term.py index 320d93c..b73d028 100644 --- a/creator/memory/long_term.py +++ b/creator/memory/long_term.py @@ -1,6 +1,6 @@ import numpy as np -from creator.llm.llm_creator import create_embedding +from creator.llm import create_embedding from creator.retrivever.score_functions import cosine_similarity from langchain.memory.chat_message_histories import SQLChatMessageHistory diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index cff6a21..e3139d1 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -12,6 +12,7 @@ from .valid_code import is_valid_code, is_expression from .tips_utils import remove_tips from .time_utils import get_local_time +from .runnable_decorator import runnable __all__ = [ @@ -29,5 +30,6 @@ "is_valid_code", "is_expression", "remove_tips", - "get_local_time" + "get_local_time", + "runnable" ] diff --git a/creator/utils/install_command.py b/creator/utils/install_command.py index c9e5894..b5baa14 100644 --- a/creator/utils/install_command.py +++ b/creator/utils/install_command.py @@ -15,7 +15,7 @@ def generate_install_command(language: str, dependencies): return _generate_html_install_command(dependencies) else: raise NotImplementedError - + def _generate_python_install_command(dependencies): shell_command_str = 'pip show {package_name} || pip install "{package_name}' diff --git a/creator/utils/printer.py b/creator/utils/printer.py index e325416..895b7aa 100644 --- a/creator/utils/printer.py +++ b/creator/utils/printer.py @@ -2,10 +2,10 @@ import sys from rich.markdown import Markdown from rich.console import Console -from rich import print as rich_print from rich.json import JSON import io + # Save the original print function original_print = print diff --git a/creator/utils/runnable_decorator.py b/creator/utils/runnable_decorator.py new file mode 100644 index 0000000..2a46faf --- /dev/null +++ b/creator/utils/runnable_decorator.py @@ -0,0 +1,7 @@ +from langchain.schema.runnable import RunnableLambda + + +def runnable(run_name): + def decorator(func): + return RunnableLambda(func).with_config({"run_name": run_name}) + return decorator diff --git a/creator/utils/skill_doc.py b/creator/utils/skill_doc.py index aef4f57..e35c97c 100644 --- a/creator/utils/skill_doc.py +++ b/creator/utils/skill_doc.py @@ -39,4 +39,3 @@ def format_return(ret): doc += format_return(skill.skill_return) + "\n" return doc.strip() - From 0c1afeba59bf103d2afe00b8c491c1b4b77287a4 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 16:12:18 +0800 Subject: [PATCH 32/56] fix: langsmith test ok --- creator/code_interpreter/__init__.py | 5 +- creator/code_interpreter/base.py | 1 - creator/core/core.py | 9 +- creator/core/runnable.py | 120 +++++++++++++-------------- creator/core/skill.py | 5 +- 5 files changed, 69 insertions(+), 71 deletions(-) diff --git a/creator/code_interpreter/__init__.py b/creator/code_interpreter/__init__.py index f06abbc..51aca45 100644 --- a/creator/code_interpreter/__init__.py +++ b/creator/code_interpreter/__init__.py @@ -79,7 +79,10 @@ def _run( if language not in self.interpreters: self.add_interpreter(language=language) code = self.clean_code(code) - result = self.interpreters[language].run(code) + if isinstance(self.interpreters[language], SafePythonInterpreter): + result = self.interpreters[language].run(code, callbacks=run_manager.get_child() if run_manager else None) + else: + result = self.interpreters[language].run(code) self.run_history[language].append({ "code": code, "result": result diff --git a/creator/code_interpreter/base.py b/creator/code_interpreter/base.py index 4576a3f..cbef01b 100644 --- a/creator/code_interpreter/base.py +++ b/creator/code_interpreter/base.py @@ -55,7 +55,6 @@ def handle_stream_output(self, stream, is_stderr): while not self.done.is_set(): for key, _ in sel.select(timeout=0.1): # Non-blocking with a small timeout line = key.fileobj.readline() - if self.detect_program_end(line): self.done.set() break diff --git a/creator/core/core.py b/creator/core/core.py index 5a8d65f..550a3ad 100644 --- a/creator/core/core.py +++ b/creator/core/core.py @@ -48,7 +48,6 @@ def wrapper(cls, *args, **kwargs): if not can_construct_skill: print(f"[red]Warning[/red]: [yellow]Only one parameter can be provided. You provided: {provided_params}[/yellow]") return None - # Return the original function with the validated parameters return func(cls, **kwargs) return wrapper @@ -131,11 +130,11 @@ def create( file_path: Optional[str] = None, huggingface_repo_id: Optional[str] = None, huggingface_skill_path: Optional[str] = None, + save: bool = False, ) -> CodeSkill: """Main method to create a new skill.""" skill_json = None - if skill_path: skill_json_path = os.path.join(skill_path, "skill.json") return cls._create_from_skill_json_path(skill_json_path) @@ -144,7 +143,7 @@ def create( return cls._create_from_skill_json_path(skill_json_path) if request: - skill_json = create_skill_from_request.invoke({"request": request}) + skill_json = create_skill_from_request(request) if messages_json_path: with open(messages_json_path, encoding="utf-8") as f: @@ -155,10 +154,10 @@ def create( file_content = "### file name: " + os.path.basename(file_path) + "\n---" + f.read() if messages is not None and len(messages) > 0: - skill_json = create_skill_from_messages.invoke({"messages": messages}) + skill_json = create_skill_from_messages(messages) elif file_content is not None: - skill_json = create_skill_from_file_content.invoke({"file_content": file_content}) + skill_json = create_skill_from_file_content(file_content) elif huggingface_repo_id and huggingface_skill_path: # huggingface_skill_path pattern username/skill_name_{version}, the version is optional and default to 1.0.0 diff --git a/creator/core/runnable.py b/creator/core/runnable.py index 9183741..af45339 100644 --- a/creator/core/runnable.py +++ b/creator/core/runnable.py @@ -1,8 +1,10 @@ import json +from langchain.schema.runnable import RunnableConfig + from creator.utils import runnable, print from creator.utils import generate_install_command -from creator.config.library import config +from creator.config.library import config as creator_config from creator.llm import create_llm from creator.agents import ( create_skill_extractor_agent, @@ -26,56 +28,54 @@ def construct_create_skill_messages(request): } -@runnable(run_name="CreateSkillFromMessages") def create_skill_from_messages(messages): - skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) - skill_json = skill_extractor_agent.invoke(input={"messages": messages})["extracted_skill"] - return skill_json + skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) + return skill_extractor_agent.with_config({"run_name": "CreateSkillFromMessages"}).invoke(input={"messages": messages})["extracted_skill"] -@runnable(run_name="CreateSkillFromRequest") def create_skill_from_request(request): - config.use_rich = False - prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(config)) - config.use_rich = True - code_interpreter_agent = create_code_interpreter_agent(create_llm(config)) - skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) + creator_config.use_rich = False + prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(creator_config)) + creator_config.use_rich = True + code_interpreter_agent = create_code_interpreter_agent(create_llm(creator_config)) + skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) chain = construct_create_skill_messages | prompt_enhancer_agent | construct_create_skill_messages | code_interpreter_agent | skill_extractor_agent - skill_json = chain.invoke(input={"request": request})["extracted_skill"] + skill_json = chain.with_config({"run_name": "CreateSkillFromRequest"}).invoke(input=request)["extracted_skill"] return skill_json -@runnable(run_name="CreateSkillFromFileContent") def create_skill_from_file_content(file_content): - skill_extractor_agent = create_skill_extractor_agent(create_llm(config)) + skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) chain = construct_create_skill_messages | skill_extractor_agent - skill_json = chain.invoke(input={"request": file_content})["extracted_skill"] + skill_json = chain.with_config({"run_name": "CreateSkillFromFileContent"}).invoke(input=file_content)["extracted_skill"] return skill_json @runnable(run_name="ConstructCreatorMessages") -def _generate_install_command(language: str, dependencies): - install_script = generate_install_command(language, dependencies) - return {"install_script": install_script} +def _generate_install_command(inputs): + install_script = generate_install_command(**inputs) + return install_script @runnable(run_name="InstallSkill") -def install_skill(skill_dependencies, skill_program_language): +def install_skill(inputs, config: RunnableConfig): + skill_dependencies, skill_program_language = inputs["skill_dependencies"], inputs["skill_program_language"] if skill_dependencies is None: - return + return inputs try: - install_script = _generate_install_command.invoke({"language": skill_program_language, "dependencies": skill_dependencies}) + install_script = _generate_install_command.invoke({"language": skill_program_language, "dependencies": skill_dependencies}, config) print("> Installing dependencies", print_type="markdown") print(f"```bash\n{install_script}\n```\n", print_type="markdown") - result = config.code_interpreter.run({"language": "shell", "code": install_script}, run_name="InstallDependencies") + result = creator_config.code_interpreter.run({"language": "shell", "code": install_script}, run_name="InstallDependencies", callbacks=config.get("callbacks", None)) print(f"> Install dependencies result: {result}", print_type="markdown") except Exception as e: print(f"> Error when installing dependencies: {e}", print_type="markdown") - return + return inputs @runnable(run_name="ConstructRunSkillMessages") -def construct_run_skill_messages(skill_name, skill_program_language, skill_code, tool_result, params): +def construct_run_skill_messages(inputs): + skill_name, skill_program_language, skill_code, tool_result, params = inputs["skill_name"], inputs["skill_program_language"], inputs["skill_code"], inputs["tool_result"], inputs["params"] messages = [ {"role": "assistant", "content": "ok I will run your code", "function_call": { "name": skill_name, @@ -89,27 +89,27 @@ def construct_run_skill_messages(skill_name, skill_program_language, skill_code, @runnable(run_name="SetupSkill") -def setup_skill(skill_program_language, skill_code): - tool_result = config.code_interpreter.run({ - "language": skill_program_language, - "code": skill_code - }) - return {"tool_result": tool_result} +def setup_skill(inputs, config: RunnableConfig): + language, code = inputs["language"], inputs["code"] + tool_result = creator_config.code_interpreter.invoke({"language": language, "code": code}, config) + inputs["tool_result"] = tool_result + return inputs @runnable(run_name="RunSkill") -def run_skill(params, skill_name, skill_program_language, skill_code, skill_dependencies): - install_skill.invoke({"skill_dependencies": skill_dependencies, "skill_program_language": skill_program_language}) - tool_result = setup_skill.invoke({"skill_program_language": skill_program_language, "skill_code": skill_code})["tool_result"] - code_interpreter_agent = create_code_interpreter_agent(create_llm(config)) - code_interpreter_agent.tools[0] = config.code_interpreter - messages_inputs = construct_run_skill_messages.invoke({"skill_name": skill_name, "skill_program_language": skill_program_language, "skill_code": skill_code, "tool_result": tool_result, "params": params}) - messages = code_interpreter_agent.invoke(messages_inputs)["messages"] +def run_skill(inputs, config: RunnableConfig): + code_interpreter_agent = create_code_interpreter_agent(create_llm(config=creator_config)) + code_interpreter_agent.tools[0] = creator_config.code_interpreter + tool_inputs = {"language": inputs["skill_program_language"], "code": inputs["skill_code"]} + inputs.update(tool_inputs) + chain = (install_skill | setup_skill | construct_run_skill_messages | code_interpreter_agent).with_config({"run_name": "Steps"}) + messages = chain.invoke(inputs, config)["messages"] return messages @runnable(run_name="ConstructTestSkillMessages") -def construct_test_skill_messages(skill_repr, tool_input, tool_result): +def construct_test_skill_messages(inputs): + skill_repr, tool_input, tool_result = inputs["skill_repr"], inputs["tool_input"], inputs["tool_result"] messages = [ {"role": "user", "content": skill_repr}, {"role": "assistant", "content": "", "function_call": {"name": "run_code", "arguments": json.dumps(tool_input)}}, @@ -120,8 +120,9 @@ def construct_test_skill_messages(skill_repr, tool_input, tool_result): @runnable(run_name="TestSkill") -def test_skill(skill_program_language, skill_dependencies, skill_code, skill_repr): - install_skill.invoke({"skill_dependencies": skill_dependencies, "skill_program_language": skill_program_language}) +def test_skill(inputs, config: RunnableConfig): + code_tester_agent = create_code_tester_agent(create_llm(creator_config)) + code_tester_agent.tools[0] = creator_config.code_interpreter code = f"""\n\n import io import unittest @@ -129,46 +130,40 @@ def test_skill(skill_program_language, skill_dependencies, skill_code, skill_rep runner = unittest.TextTestRunner(stream=stream) -{skill_code} +{inputs["skill_code"]} """ - tool_inputs = {"skill_program_language": skill_program_language, "skill_code": code} - tool_result = setup_skill.invoke(tool_inputs)["tool_result"] - messages_inputs = construct_test_skill_messages.invoke({"skill_repr": skill_repr, "tool_input": tool_inputs, "tool_result": tool_result}) - code_tester_agent = create_code_tester_agent(create_llm(config)) - code_tester_agent.tools[0] = config.code_interpreter - test_result = code_tester_agent.invoke(messages_inputs)["messages"] + tool_inputs = {"language": inputs["skill_program_language"], "code": code} + inputs.update(tool_inputs) + inputs["tool_input"] = tool_inputs + chain = (install_skill | setup_skill | construct_test_skill_messages | code_tester_agent).with_config({"run_name": "Steps"}) + test_result = chain.invoke(inputs, config)["output"] return test_result @runnable(run_name="ConstructRefactorSkillMessages") -def construct_refactor_skill_messages(conversation_history, refactor_type, skill_repr, skill_program_language, skill_code, user_request): +def construct_refactor_skill_messages(inputs): + conversation_history, refactor_type, skill_repr, skill_program_language, skill_code, user_request = inputs["conversation_history"], inputs["refactor_type"], inputs["skill_repr"], inputs["skill_program_language"], inputs["skill_code"], inputs["user_request"] messages = [ {"role": "system", "content": f"Your action type is: {refactor_type}"}, {"role": "function", "name": "show_skill", "content": skill_repr}, {"role": "function", "name": "show_code", "content": f"current skill code:\n```{skill_program_language}\n{skill_code}\n```"}, - {"role": "user", "content": "{user_request}\nplease output only one skill object" if refactor_type in ("Combine", "Refine") else "\nplease help me decompose the skill object into different independent skill objects"} + {"role": "user", "content": f"{user_request}\nplease output only one skill object" if refactor_type in ("Combine", "Refine") else "\nplease help me decompose the skill object into different independent skill objects"} ] messages = conversation_history + [{"role": "system", "content": "Above context is conversation history from other agents. Now let's refactor our skill."}] + messages return {"messages": messages} @runnable(run_name="RefactorSkill") -def refactor_skill(conversation_history, refactor_type, skill_program_language, skill_code, skill_repr, user_request): - code_refactor_agent = create_code_refactor_agent(create_llm(config)) +def refactor_skill(inputs, config: RunnableConfig): + code_refactor_agent = create_code_refactor_agent(create_llm(creator_config)) chain = construct_refactor_skill_messages | code_refactor_agent - refactored_skill_jsons = chain.invoke({ - "conversation_history": conversation_history, - "refactor_type": refactor_type, - "skill_repr": skill_repr, - "skill_program_language": skill_program_language, - "skill_code": skill_code, - "user_request": user_request - })["refacted_skills"] + refactored_skill_jsons = chain.invoke(inputs, config)["refacted_skills"] return refactored_skill_jsons @runnable(run_name="AutoOptimizeSkill") -def auto_optimize_skill(old_skill, retry_times): +def auto_optimize_skill(inputs, config: RunnableConfig): + old_skill, retry_times = inputs["old_skill"], inputs["retry_times"] skill = old_skill.model_copy(deep=True) refined = False conversation_history = [] if skill.conversation_history is None else skill.conversation_history @@ -178,8 +173,8 @@ def auto_optimize_skill(old_skill, retry_times): "skill_program_language": skill.skill_program_language, "skill_dependencies": skill.skill_dependencies, "skill_code": skill.skill_code, - "skill_repr": skill.skill_repr - }) + "skill_repr": repr(skill) + }, config) conversation_history = conversation_history + test_result["messages"] if "test_summary" in test_result: test_summary = test_result["test_summary"] @@ -191,6 +186,7 @@ def auto_optimize_skill(old_skill, retry_times): "test_summary": test_summary, } print(f"> Auto Refine Skill {i+1}/{retry_times}", print_type="markdown") + skill.Config.runnable_config = config skill = skill > "I have tested the skill, but it failed, please refine it." skill.conversation_history = conversation_history if all_passed: diff --git a/creator/core/skill.py b/creator/core/skill.py index f31a534..286a98f 100644 --- a/creator/core/skill.py +++ b/creator/core/skill.py @@ -148,6 +148,7 @@ class Config: skills_to_combine = [] user_request = "please help me refine the skill object" refactor_type = "Refine" + runnable_config = None def __init__(self, **data): super().__init__(**data) @@ -250,7 +251,7 @@ def __gt__(self, user_request:str): return self.refactor() def run(self, inputs: Union[str, dict[str, Any]]): - return run_skill.invoke({"params": inputs, "skill_name": self.skill_name, "skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "skill_dependencies": self.skill_dependencies}) + return run_skill.invoke(input={"params": inputs, "skill_name": self.skill_name, "skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "skill_dependencies": self.skill_dependencies}) def test(self): if self.conversation_history is None: @@ -278,7 +279,7 @@ def refactor(self): "skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "user_request": self.Config.user_request, - }) + }, self.Config.runnable_config) refactored_skills = [] for refactored_skill_json in refactored_skill_jsons: refactored_skill = CodeSkill(**refactored_skill_json) From 01c6b5412dd915ed152374cda6adaab2b999fc1d Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 16:29:39 +0800 Subject: [PATCH 33/56] feat: auto generate run url --- creator/core/runnable.py | 7 ++++++- creator/core/skill.py | 6 +++++- creator/utils/__init__.py | 5 +++-- creator/utils/runnable_decorator.py | 13 +++++++++++++ 4 files changed, 27 insertions(+), 4 deletions(-) diff --git a/creator/core/runnable.py b/creator/core/runnable.py index af45339..1dd152f 100644 --- a/creator/core/runnable.py +++ b/creator/core/runnable.py @@ -2,7 +2,7 @@ from langchain.schema.runnable import RunnableConfig -from creator.utils import runnable, print +from creator.utils import runnable, print, print_run_url from creator.utils import generate_install_command from creator.config.library import config as creator_config from creator.llm import create_llm @@ -28,11 +28,13 @@ def construct_create_skill_messages(request): } +@print_run_url def create_skill_from_messages(messages): skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) return skill_extractor_agent.with_config({"run_name": "CreateSkillFromMessages"}).invoke(input={"messages": messages})["extracted_skill"] +@print_run_url def create_skill_from_request(request): creator_config.use_rich = False prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(creator_config)) @@ -44,6 +46,7 @@ def create_skill_from_request(request): return skill_json +@print_run_url def create_skill_from_file_content(file_content): skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) chain = construct_create_skill_messages | skill_extractor_agent @@ -119,6 +122,7 @@ def construct_test_skill_messages(inputs): return {"messages": messages} +@print_run_url @runnable(run_name="TestSkill") def test_skill(inputs, config: RunnableConfig): code_tester_agent = create_code_tester_agent(create_llm(creator_config)) @@ -153,6 +157,7 @@ def construct_refactor_skill_messages(inputs): return {"messages": messages} +@print_run_url @runnable(run_name="RefactorSkill") def refactor_skill(inputs, config: RunnableConfig): code_refactor_agent = create_code_refactor_agent(create_llm(creator_config)) diff --git a/creator/core/skill.py b/creator/core/skill.py index 286a98f..c9009f7 100644 --- a/creator/core/skill.py +++ b/creator/core/skill.py @@ -2,7 +2,7 @@ from typing import List, Dict, Optional, Union, Any from datetime import datetime -from creator.utils import remove_title +from creator.utils import remove_title, print_run_url from creator.config.library import config from creator.utils import generate_skill_doc, generate_install_command, print, generate_language_suffix from creator.hub.huggingface import hf_repo_update, hf_push @@ -250,9 +250,11 @@ def __gt__(self, user_request:str): self.Config.refactor_type = "Combine" return self.refactor() + @print_run_url def run(self, inputs: Union[str, dict[str, Any]]): return run_skill.invoke(input={"params": inputs, "skill_name": self.skill_name, "skill_program_language": self.skill_program_language, "skill_code": self.skill_code, "skill_dependencies": self.skill_dependencies}) + @print_run_url def test(self): if self.conversation_history is None: self.conversation_history = [] @@ -265,6 +267,7 @@ def test(self): self.conversation_history = self.conversation_history + test_result["messages"] return self.test_summary + @print_run_url def refactor(self): if self.conversation_history is None: self.conversation_history = [] @@ -291,6 +294,7 @@ def refactor(self): return refactored_skills[0] return refactored_skills + @print_run_url def auto_optimize(self, retry_times=3): optimized_result = auto_optimize_skill.invoke({"old_skill": self, "retry_times": retry_times}) test_summary = optimized_result["test_summary"] diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index e3139d1..5e7a0d2 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -12,7 +12,7 @@ from .valid_code import is_valid_code, is_expression from .tips_utils import remove_tips from .time_utils import get_local_time -from .runnable_decorator import runnable +from .runnable_decorator import runnable, print_run_url __all__ = [ @@ -31,5 +31,6 @@ "is_expression", "remove_tips", "get_local_time", - "runnable" + "runnable", + "print_run_url" ] diff --git a/creator/utils/runnable_decorator.py b/creator/utils/runnable_decorator.py index 2a46faf..ef52efb 100644 --- a/creator/utils/runnable_decorator.py +++ b/creator/utils/runnable_decorator.py @@ -1,7 +1,20 @@ from langchain.schema.runnable import RunnableLambda +from langchain.callbacks import tracing_v2_enabled +from .printer import print def runnable(run_name): def decorator(func): return RunnableLambda(func).with_config({"run_name": run_name}) return decorator + + +def print_run_url(func): + def wrapper(*args, **kwargs): + with tracing_v2_enabled() as cb: + result = func(*args, **kwargs) + run_url = cb.get_run_url() + if run_url is not None: + print(f"Langsmith Run URL: [click]({run_url})", print_type="markdown") + return result + return wrapper From 2def92b2820fb8ce7f548e81b7e4ca4d26e3a8ea Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 17:09:09 +0800 Subject: [PATCH 34/56] feat: add langsmith status check --- creator/utils/langsmith_utils.py | 21 +++++++++++++++++++++ creator/utils/runnable_decorator.py | 12 ++++++++---- pyproject.toml | 2 +- 3 files changed, 30 insertions(+), 5 deletions(-) create mode 100644 creator/utils/langsmith_utils.py diff --git a/creator/utils/langsmith_utils.py b/creator/utils/langsmith_utils.py new file mode 100644 index 0000000..4f60508 --- /dev/null +++ b/creator/utils/langsmith_utils.py @@ -0,0 +1,21 @@ +from langsmith import Client +from langsmith.utils import LangSmithError +from .printer import print + + +def check_langsmith_ok(): + cli = Client() + ok = False + try: + cli.read_project(project_name="open-creator") + except LangSmithError as e: + if "Bad Gateway" in str(e): + print("[red]Warning:[/red] [yellow]Langsmith is not running. Please run `langsmith start`.[/yellow]") + return ok + else: + cli.create_project(project_name="open-creator") + ok = True + return ok + + +langsmith_ok = check_langsmith_ok() diff --git a/creator/utils/runnable_decorator.py b/creator/utils/runnable_decorator.py index ef52efb..d4afc23 100644 --- a/creator/utils/runnable_decorator.py +++ b/creator/utils/runnable_decorator.py @@ -1,6 +1,7 @@ from langchain.schema.runnable import RunnableLambda from langchain.callbacks import tracing_v2_enabled from .printer import print +from .langsmith_utils import langsmith_ok def runnable(run_name): @@ -11,10 +12,13 @@ def decorator(func): def print_run_url(func): def wrapper(*args, **kwargs): - with tracing_v2_enabled() as cb: + if langsmith_ok: + with tracing_v2_enabled() as cb: + result = func(*args, **kwargs) + run_url = cb.get_run_url() + if run_url is not None: + print(f"Langsmith Run URL: [{run_url}]{run_url}", print_type="markdown") + else: result = func(*args, **kwargs) - run_url = cb.get_run_url() - if run_url is not None: - print(f"Langsmith Run URL: [click]({run_url})", print_type="markdown") return result return wrapper diff --git a/pyproject.toml b/pyproject.toml index 9ad6e84..3400bde 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -28,7 +28,7 @@ fastapi = "^0.103.1" uvicorn = "^0.23.2" streamlit = "^1.27.2" questionary = "^2.0.1" - +langsmith = "^0.0.43" [tool.poetry.dependencies.pyreadline3] version = "^3.4.1" From 07828e70c59252987f8a4529952f2b0cc2f50f49 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 17:20:37 +0800 Subject: [PATCH 35/56] feat: auto update user config if project config update new key --- creator/config/load_config.py | 31 ++++++++++++++++++++++--------- 1 file changed, 22 insertions(+), 9 deletions(-) diff --git a/creator/config/load_config.py b/creator/config/load_config.py index 832a347..8ea20af 100644 --- a/creator/config/load_config.py +++ b/creator/config/load_config.py @@ -11,9 +11,9 @@ def load_yaml_config(): """ Load the configuration from a YAML file. - + If the config file doesn't exist in the user's directory, it copies the default from the project directory. - + Returns: dict: The loaded YAML configuration. """ @@ -35,17 +35,30 @@ def load_yaml_config(): if not os.path.exists(user_config_path): shutil.copy(project_config_path, user_config_path) - # Load YAML config file using the new path + # Load YAML config files + with open(project_config_path, mode='r', encoding="utf-8") as f: + project_config = yaml.safe_load(f) + with open(user_config_path, mode='r', encoding="utf-8") as f: - yaml_config = yaml.safe_load(f) - + user_config = yaml.safe_load(f) + + updated = False + for key, value in project_config.items(): + if key not in user_config: + user_config[key] = value + updated = True + + # If user config was updated, write back to file + if updated: + with open(user_config_path, mode='w', encoding="utf-8") as f: + yaml.dump(user_config, f) + # env vs yaml, yaml first if not empty - for key, value in yaml_config.items(): + for key, value in user_config.items(): if os.environ.get(key) and not value: - yaml_config[key] = os.environ.get(key) + user_config[key] = os.environ.get(key) # if yaml has some configs that env does not, write to env if not os.environ.get(key) and value: os.environ[key] = str(value) - return yaml_config - + return user_config From 3f4efa9f6fe7577031580976c2043b0f35a37bee Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Thu, 26 Oct 2023 17:42:34 +0800 Subject: [PATCH 36/56] feat: update langchain dependancy and add add_to_memory interface --- creator/agents/base.py | 5 +++++ pyproject.toml | 2 +- 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/creator/agents/base.py b/creator/agents/base.py index ad2715f..85bde03 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -110,6 +110,10 @@ def messages_hot_fix(self, langchain_messages): def preprocess_inputs(self, inputs: Dict[str, Any]): return inputs + def add_to_memory(self, messages): + """Add message to long-term memory""" + pass + def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: run_manager_callbacks = run_manager.get_child() if run_manager else None inputs = self.preprocess_inputs(inputs) @@ -139,6 +143,7 @@ def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackMan current_try += 1 self.end_callbacks(message) langchain_messages = remove_tips(langchain_messages) + self.add_to_memory(langchain_messages) openai_messages = list(map(convert_message_to_dict, langchain_messages)) return openai_messages diff --git a/pyproject.toml b/pyproject.toml index 3400bde..35d5428 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -12,7 +12,7 @@ include = ["creator/config.yaml"] [tool.poetry.dependencies] python = "^3.10" rich = "^13.5.2" -langchain = ">=0.0.317" +langchain = ">=0.0.323" huggingface_hub = "^0.17.2" loguru = "^0.7.2" pydantic = "^2.0.3" From df186564831fc42b9bcdc3998d2f606a00fe1b78 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 27 Oct 2023 12:13:27 +0800 Subject: [PATCH 37/56] feat: add token count for function call and remove long_term_memory support for base agent --- creator/agents/base.py | 19 +- creator/agents/memgpt_agent.py | 262 ++++++++++++++++++--------- creator/llm/chatllm_with_trim.py | 4 +- creator/llm/format_function_calls.py | 93 ++++++++++ creator/llm/tokentrim.py | 14 +- creator/memory/schema.py | 30 ++- 6 files changed, 310 insertions(+), 112 deletions(-) create mode 100644 creator/llm/format_function_calls.py diff --git a/creator/agents/base.py b/creator/agents/base.py index 85bde03..55a2abc 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -14,8 +14,6 @@ from creator.utils import get_user_info, ask_run_code_confirm, remove_tips from creator.callbacks.buffer_manager import buffer_output_manager -from creator.memory.schema import MessageConverter -from langchain.memory.chat_message_histories import SQLChatMessageHistory from creator.config.library import config @@ -27,7 +25,6 @@ class BaseAgent(LLMChain): system_template: str = "" allow_user_confirm: bool = False prompt: ChatPromptTemplate = ChatPromptTemplate.from_messages(messages=["system", ""]) - long_term_memory: SQLChatMessageHistory = None @property def _chain_type(self): @@ -37,15 +34,6 @@ def _chain_type(self): def input_keys(self) -> List[str]: return ["messages"] - def build_memory(self, session_id=None): - if session_id is None: - session_id = str(uuid.uuid4()) - self.long_term_memory = SQLChatMessageHistory( - session_id=session_id, - connection_string=f"sqlite:///{config.memory_path}/.langchain.db", - custom_message_converter=MessageConverter() - ) - def construct_prompt(self, langchain_messages: Dict[str, Any]): prompt = ChatPromptTemplate.from_messages(messages=[ ("system", self.system_template + get_user_info()), @@ -96,6 +84,9 @@ def run_tool(self, function_call: Dict[str, Any], run_manager: Optional[Callback tool_result = FunctionMessage(name=function_name, content=tool_result) self.update_tool_result_in_callbacks(tool_result) break + if tool_result is None: + tool_names = list(map(lambda x: x.name, self.tools)) + tool_result = FunctionMessage(name="system_alert", content=f"Unknown function name: {function_name}. Available functions: {tool_names}") return tool_result def human_confirm(self): @@ -117,10 +108,9 @@ def add_to_memory(self, messages): def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: run_manager_callbacks = run_manager.get_child() if run_manager else None inputs = self.preprocess_inputs(inputs) - session_id = inputs.get("session_id", None) - self.build_memory(session_id) messages = inputs.pop("messages") langchain_messages = convert_openai_messages(messages) + self.llm.function_calls = self.function_schemas llm_with_functions = self.llm.bind(functions=self.function_schemas) current_try = 0 while current_try < self.total_tries: @@ -143,7 +133,6 @@ def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackMan current_try += 1 self.end_callbacks(message) langchain_messages = remove_tips(langchain_messages) - self.add_to_memory(langchain_messages) openai_messages = list(map(convert_message_to_dict, langchain_messages)) return openai_messages diff --git a/creator/agents/memgpt_agent.py b/creator/agents/memgpt_agent.py index 445cdbb..3511451 100644 --- a/creator/agents/memgpt_agent.py +++ b/creator/agents/memgpt_agent.py @@ -1,64 +1,29 @@ from typing import List, Dict, Any, Optional import datetime import json +import uuid -from langchain.schema.messages import SystemMessage, AIMessage, FunctionMessage +from langchain.schema.messages import FunctionMessage from langchain.output_parsers.json import parse_partial_json +from langchain.callbacks.manager import CallbackManager from langchain.prompts import ChatPromptTemplate -from creator.code_interpreter import CodeInterpreter, language_map -from langchain.adapters.openai import convert_message_to_dict, convert_openai_messages +from langchain.adapters.openai import convert_openai_messages, convert_message_to_dict from creator.config.library import config -from creator.utils import load_system_prompt -from creator.llm import create_llm -from langchain.tools import StructuredTool -from creator.utils import get_user_info, get_local_time +from creator.utils import load_system_prompt, get_user_info from creator.callbacks.streaming_stdout import OutputBufferStreamingHandler - from creator.agents.base import BaseAgent - - -def package_function_response(was_success, response_string, timestamp=None): - - formatted_time = get_local_time() if timestamp is None else timestamp - packaged_message = { - "status": 'OK' if was_success else 'Failed', - "message": response_string, - "time": formatted_time, - } - return json.dumps(packaged_message) - - -def get_initial_boot_messages(): - return [ - # first message includes both inner monologue and function call to send_message - { - "role": "assistant", - "content": "Bootup sequence complete. Persona activated. Testing messaging functionality.", - "function_call": { - "name": "send_message", - "arguments": "{\n \"message\": \"" + "More human than human is our motto." + "\"\n}" - } - }, - # obligatory function return message - { - "role": "function", - "name": "send_message", - "content": package_function_response(True, None) - } - ] - - -def get_login_event_message(last_login='Never (first login)', include_location=False, location_name='San Francisco, CA, USA'): - # Package the message with time and location - formatted_time = get_local_time() - packaged_message = { - "type": 'login', - "last_login": last_login, - "time": formatted_time, - } - if include_location: - packaged_message["location"] = location_name - return {"role": "user", "content": json.dumps(packaged_message)}, +from creator.utils.memgpt_utils import ( + get_initial_boot_messages, + get_login_event, + DEFAULT_PERSONA, + DEFAULT_HUMAN, + DEFAULT_AGENT_SUBTASKS, + package_message, + MESSAGE_SUMMARY_WARNING_TOKENS, + MESSAGE_SUMMARY_WARNING_STR +) +from creator.memory.schema import MessageConverter, ChatMessageHistory +from creator.llm.tokentrim import num_tokens_from_messages class MemGPTAgent(BaseAgent): @@ -66,11 +31,27 @@ class MemGPTAgent(BaseAgent): allow_user_confirm: bool = config.run_human_confirm subagent: Optional[BaseAgent] = None pause_heartbeats_start: Optional[datetime.datetime] = None + is_new_session: bool = True + heartbeat_request: bool = False + long_term_memory: ChatMessageHistory = None @property def _chain_type(self): return "MemGPTAgent" + @property + def input_keys(self) -> List[str]: + return ["user_request"] + + def build_memory(self, session_id=None): + if session_id is None: + session_id = str(uuid.uuid4()) + self.long_term_memory = ChatMessageHistory( + session_id=session_id, + connection_string=f"sqlite:///{config.memory_path}/.langchain.db", + custom_message_converter=MessageConverter() + ) + async def send_message(self, message: str, receiver: str = "human"): """Sends a message to a specified receiver""" if receiver == "human": @@ -98,18 +79,6 @@ async def search_memory(self, memory_type: str, page: int = 0, query: str = "", """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" pass - def construct_prompt(self, langchain_messages: Dict[str, Any]): - init_messages = convert_openai_messages(get_initial_boot_messages()) - login_message = convert_openai_messages(get_login_event_message()) - - prompt = ChatPromptTemplate.from_messages(messages=[ - ("system", self.system_template + get_user_info()), - *init_messages, - *login_message, - *langchain_messages - ]) - return prompt - def get_callbacks(self): callbacks = [] if self.llm.callbacks is not None: @@ -120,35 +89,148 @@ def get_callbacks(self): return callbacks def preprocess_inputs(self, inputs: Dict[str, Any]): - inputs["memory_edit_timestamp"] = 0 - inputs["recall_memory_count"] = 0 - inputs["archival_memory_count"] = 0 - inputs["persona"] = """ -The following is a blank slate starter persona, I need to expand this to develop my own personality. - -My name is MemGPT. -I am kind, thoughtful, and inquisitive. -""" - inputs["human"] = """ -This is what I know so far about the user, I should expand this as I learn more about them. - -First name: Chad -Last name: ? -Gender: Male -Age: ? -Nationality: ? -Occupation: Computer science PhD student at UC Berkeley -Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO -""" - inputs["subagent_tasks"] = """ -- create/save/search skill -- run/test/refactor skill -- show skill -""" + """Preprocesses inputs to the agent""" + session_id = inputs.get("session_id", None) + if session_id is None: + session_id = str(uuid.uuid4()) + inputs["session_id"] = session_id + self.is_new_session = True + if self.long_term_memory is None: + self.build_memory(session_id=session_id) + + inputs["memory_edit_timestamp"] = self.long_term_memory.memory_edit_timestamp + inputs["recall_memory_count"] = self.long_term_memory.recall_memory_count + inputs["archival_memory_count"] = self.long_term_memory.archival_memory_count + if inputs["recall_memory_count"] > 0: + self.is_new_session = False + + if "persona" not in inputs: + inputs["persona"] = DEFAULT_PERSONA + if "human" not in inputs: + inputs["human"] = DEFAULT_HUMAN + if "subagent_tasks" not in inputs: + inputs["subagent_tasks"] = DEFAULT_AGENT_SUBTASKS + return inputs + def construct_prompt(self, langchain_messages: Dict[str, Any]): + extra_messages = [] + if self.is_first: + init_messages = convert_openai_messages(get_initial_boot_messages()) + login_message = convert_openai_messages(get_login_event()) + extra_messages = init_messages + login_message + + prompt = ChatPromptTemplate.from_messages(messages=[ + ("system", self.system_template + get_user_info()), + *extra_messages, + *langchain_messages + ]) + return prompt + + + async def arun_tool(self, function_call: Dict[str, Any], run_manager: Optional[CallbackManager] = None): + function_name = function_call.get("name", "") + + available_functions = { + "pause_heartbeats": self.pause_heartbeats, + "send_message": self.send_message, + "add_memory": self.add_memory, + "modify_memory": self.modify_memory, + "search_memory": self.search_memory, + } -def create_memGPT_agent(llm, subagent): + # Failure case 1: function name is wrong + try: + function_to_call = available_functions[function_name] + except KeyError: + extra_info = {"status": 'Failed', "message": f'No function named {function_name}'} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # Failure case 2: function name is OK, but function args are bad JSON + try: + raw_function_args = function_call.get("arguments", "{}") + function_args = parse_partial_json(raw_function_args) + except Exception: + extra_info = {"status": 'Failed', "message": f"Error parsing JSON for function '{function_name}' arguments: {raw_function_args}"} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # (Still parsing function args) + # Handle requests for immediate heartbeat + heartbeat_request = function_args.pop('request_heartbeat', False) + if not (isinstance(heartbeat_request, bool) or heartbeat_request is None): + print(f"> Warning: 'request_heartbeat' arg parsed was not a bool or None, type={type(heartbeat_request)}, value={heartbeat_request}", print_type="markdown") + heartbeat_request = False + self.heartbeat_request = heartbeat_request + + # Failure case 3: function failed during execution + try: + function_response_string = await function_to_call(**function_args) + extra_info = {"status": 'OK', "message": function_response_string} + function_response = package_message(message_type=None, extra_info=extra_info) + except Exception as e: + extra_info = {"status": 'Failed', "message": f"Error calling function {function_name} with args {function_args}: {str(e)}"} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # If no failures happened along the way: ... + # Step 4: send the info on the function call and function response to GPT + message = FunctionMessage(name=function_name, content=function_response) + return message + + def check_token_limit(self, langchain_messages): + """Check if the token limit is reached, if so, return the messages with a warning message""" + token_cnt = num_tokens_from_messages(convert_message_to_dict(langchain_messages), model=config.model) + if token_cnt > self.token_limit: + warning_message = package_message(message_type="system_alert", extra_info={"message": MESSAGE_SUMMARY_WARNING_STR}) + langchain_messages.append(warning_message) + + async def arun_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: + run_manager_callbacks = run_manager.get_child() if run_manager else None + inputs = self.preprocess_inputs(inputs) + # handle usre request + user_request = inputs.get("user_request") + user_message = package_message(message_type="user_message", extra_info={"message": user_request}) + self.long_term_memory.add_user_message(user_message) + langchain_messages = self.long_term_memory.messages() + + # check token limit warning + # convert_message_to_dict(langchain_messages) + + # llm_with_functions = self.llm.bind(functions=self.function_schemas) + # current_try = 0 + # while current_try < self.total_tries: + # self.start_callbacks() + # prompt = self.construct_prompt(langchain_messages) + # llm_chain = (prompt | llm_with_functions | self.postprocess_mesasge).with_config({"run_name": f"Iteration {current_try+1}"}) + # message = llm_chain.invoke(inputs, {"callbacks": run_manager_callbacks}) + # langchain_messages.append(message) + # function_call = message.additional_kwargs.get("function_call", None) + # if function_call is None: + # self.end_callbacks(message) + # break + + # tool_result = self.run_tool(function_call, run_manager_callbacks) + # if tool_result is None: + # self.end_callbacks(message) + # break + # langchain_messages.append(tool_result) + # langchain_messages = self.messages_hot_fix(langchain_messages) + # current_try += 1 + # self.end_callbacks(message) + # langchain_messages = remove_tips(langchain_messages) + # openai_messages = list(map(convert_message_to_dict, langchain_messages)) + # return openai_messages + + +def create_memgpt_agent(llm, subagent=None): template = load_system_prompt(config.memGPT_prompt_path) with open(config.memGPT_schema_path) as f: function_schemas = json.load(f) diff --git a/creator/llm/chatllm_with_trim.py b/creator/llm/chatllm_with_trim.py index 161da29..44086b2 100644 --- a/creator/llm/chatllm_with_trim.py +++ b/creator/llm/chatllm_with_trim.py @@ -5,11 +5,13 @@ class TrimMixin: + function_calls: Optional[List[Dict]] = None + def _create_message_dicts( self, messages: List[BaseMessage], stop: Optional[List[str]] ) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]: message_dicts, params = super()._create_message_dicts(messages, stop) - message_dicts = trim(messages=message_dicts, model=self.model_name, max_tokens=self.max_tokens) + message_dicts = trim(messages=message_dicts, model=self.model_name, max_tokens=self.max_tokens, function_calls=self.function_calls) return message_dicts, params diff --git a/creator/llm/format_function_calls.py b/creator/llm/format_function_calls.py new file mode 100644 index 0000000..b014b6f --- /dev/null +++ b/creator/llm/format_function_calls.py @@ -0,0 +1,93 @@ +import json +import textwrap + + +# reference: +# https://gist.github.com/CGamesPlay/dd4f108f27e2eec145eedf5c717318f5 +# https://community.openai.com/t/how-to-calculate-the-tokens-when-using-function-call/266573/24 + +# Function overhead is 16: 3 for the system message plus this template. + +FUNCTION_HEADER_STR = """# Tools + +## functions + +namespace functions { + +} // namespace functions""" + + +def resolve_ref(schema, json_schema): + if schema.get("$ref") is not None: + ref = schema["$ref"][14:] + schema = json_schema["definitions"][ref] + return schema + + +def format_enum(schema): + return " | ".join(json.dumps(o) for o in schema["enum"]) + + +def format_default(schema): + v = schema["default"] + if schema["type"] == "number": + return f"{v:.1f}" if float(v).is_integer() else str(v) + else: + return str(v) + + +def format_object(schema, indent, json_schema): + result = "{\n" + if "properties" not in schema or len(schema["properties"]) == 0: + if schema.get("additionalProperties", False): + return "object" + return None + for key, value in schema["properties"].items(): + value = resolve_ref(value, json_schema) + value_rendered = format_schema(value, indent + 1, json_schema) + if value_rendered is None: + continue + if "description" in value and indent == 0: + for line in textwrap.dedent(value["description"]).strip().split("\n"): + result += f"{' '*indent}// {line}\n" + optional = "" if key in schema.get("required", {}) else "?" + comment = ( + "" + if value.get("default") is None + else f" // default: {format_default(value)}" + ) + result += f"{' '*indent}{key}{optional}: {value_rendered},{comment}\n" + result += (" " * (indent - 1)) + "}" + return result + + +def format_schema(schema, indent, json_schema): + schema = resolve_ref(schema, json_schema) + if "enum" in schema: + return format_enum(schema) + elif schema["type"] == "object": + return format_object(schema, indent, json_schema) + elif schema["type"] == "integer": + return "number" + elif schema["type"] in ["string", "number"]: + return schema["type"] + elif schema["type"] == "array": + return format_schema(schema["items"], indent, json_schema) + "[]" + else: + raise ValueError("unknown schema type " + schema["type"]) + + +def format_tool(tool): + json_schema = tool["parameters"] + result = f"// {tool['description']}\ntype {tool['name']} = (" + formatted = format_object(json_schema, 0, json_schema) + if formatted is not None: + result += "_: " + formatted + result += ") => any;\n\n" + return result + + +def get_function_calls_token_count(encoder, function_calls): + head_cnt = 3 + len(encoder.encode(FUNCTION_HEADER_STR)) + functions_cnt = sum(len(encoder.encode(format_tool(f))) for f in function_calls) + return head_cnt + functions_cnt diff --git a/creator/llm/tokentrim.py b/creator/llm/tokentrim.py index 9edd7cb..1199e99 100644 --- a/creator/llm/tokentrim.py +++ b/creator/llm/tokentrim.py @@ -1,5 +1,10 @@ import tiktoken from typing import List, Dict, Any, Optional +from .format_function_calls import get_function_calls_token_count + +# reference: +# https://github.com/openai/openai-cookbook/blob/main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb +# https://github.com/KillianLucas/tokentrim/blob/main/tokentrim/tokentrim.py # Define model configurations in a centralized location @@ -82,7 +87,7 @@ def tokens_for_message(message: Dict[str, Any], encoding: Any, config: Dict[str, Calculate the number of tokens for a single message. """ num_tokens = config['tokens_per_message'] - + for key, value in message.items(): try: num_tokens += len(encoding.encode(str(value))) @@ -95,7 +100,6 @@ def tokens_for_message(message: Dict[str, Any], encoding: Any, config: Dict[str, return num_tokens -# Refactored main function def num_tokens_from_messages(messages: List[Dict[str, Any]], model: Optional[str] = None) -> int: """ Function to return the number of tokens used by a list of messages. @@ -135,7 +139,8 @@ def trim( messages: List[Dict[str, Any]], model: Optional[str] = None, trim_ratio: float = 0.75, - max_tokens: Optional[int] = None + max_tokens: Optional[int] = None, + function_calls: List[Dict] = None, ) -> List[Dict[str, Any]]: """ Trim a list of messages to fit within a model's token limit. @@ -148,6 +153,9 @@ def trim( config = get_model_config(model) max_tokens = int(config['max_tokens'] * trim_ratio) + if function_calls is not None: + max_tokens -= get_function_calls_token_count(get_encoding_for_model(model), function_calls) + total_tokens = num_tokens_from_messages(messages, model) if total_tokens <= max_tokens: return messages diff --git a/creator/memory/schema.py b/creator/memory/schema.py index 7ef38a7..b25ff1f 100644 --- a/creator/memory/schema.py +++ b/creator/memory/schema.py @@ -1,10 +1,13 @@ from typing import Any +import datetime + from sqlalchemy.orm import declarative_base from sqlalchemy import Column, Integer, Text, DateTime, JSON + from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, FunctionMessage from langchain.memory.chat_message_histories.sql import BaseMessageConverter -import datetime - +from langchain.memory.chat_message_histories import SQLChatMessageHistory +from creator.utils.time_utils import get_local_time Base = declarative_base() @@ -56,7 +59,7 @@ def from_sql_model(self, sql_message: Any) -> BaseMessage: raise ValueError(f'Unknown message type: {sql_message.type}') def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: - now = datetime.now() + now = get_local_time() if isinstance(message, FunctionMessage): message.additional_kwargs.update({"name": message.name}) message.additional_kwargs.update({"created_at": now}) @@ -70,3 +73,24 @@ def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: def get_sql_model_class(self) -> Any: return MemoryMessage + + +class ChatMessageHistory(SQLChatMessageHistory): + + @property + def memory_edit_timestamp(self): + messages = self.messages() + now = get_local_time() + if messages: + return messages[-1].additional_kwargs.get("created_at", now) + return now + + @property + def recall_memory_count(self): + messages = self.messages() + return len([m for m in messages if m.type != "archival"]) + + @property + def archival_memory_count(self): + messages = self.messages() + return len([m for m in messages if m.type == "archival"]) From 2cc73b755da1cb8a65fe0fa42a7525953ce1c4c0 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Fri, 27 Oct 2023 19:40:00 +0800 Subject: [PATCH 38/56] feat: support memgpt and most of functionalities test ok --- creator/agents/__init__.py | 4 +- creator/agents/base.py | 7 +- creator/agents/creator_agent.py | 8 +- creator/agents/memgpt_agent.py | 244 ------------------ creator/app/server.py | 5 +- creator/app/streamlit_app.py | 13 +- creator/callbacks/streaming_stdout.py | 8 +- creator/client/repl/handler.py | 12 +- creator/config.yaml | 1 - creator/config/library.py | 6 +- creator/llm/chatllm_with_trim.py | 8 +- creator/llm/format_function_calls.py | 2 +- creator/llm/tokentrim.py | 12 +- creator/memgpt/__init__.py | 8 + creator/memgpt/agent.py | 230 +++++++++++++++++ creator/memgpt/config.py | 39 +++ creator/memgpt/constants.py | 38 +++ creator/memgpt/memory/__init__.py | 78 ++++++ creator/memgpt/memory/archival_memory.py | 20 ++ creator/{ => memgpt}/memory/base.py | 4 +- .../memory/core_memory.py} | 4 +- .../memory/recall_memory.py} | 33 +-- creator/{ => memgpt}/memory/schema.py | 50 ++-- creator/memgpt/message_handler.py | 52 ++++ .../prompts/function_schema.json} | 2 +- .../prompts/system_prompt.md} | 0 creator/{utils => memgpt}/time_utils.py | 0 creator/memgpt/tools_handler.py | 76 ++++++ creator/utils/__init__.py | 2 - 29 files changed, 617 insertions(+), 349 deletions(-) delete mode 100644 creator/agents/memgpt_agent.py create mode 100644 creator/memgpt/__init__.py create mode 100644 creator/memgpt/agent.py create mode 100644 creator/memgpt/config.py create mode 100644 creator/memgpt/constants.py create mode 100644 creator/memgpt/memory/__init__.py create mode 100644 creator/memgpt/memory/archival_memory.py rename creator/{ => memgpt}/memory/base.py (94%) rename creator/{memory/short_term.py => memgpt/memory/core_memory.py} (94%) rename creator/{memory/long_term.py => memgpt/memory/recall_memory.py} (69%) rename creator/{ => memgpt}/memory/schema.py (68%) create mode 100644 creator/memgpt/message_handler.py rename creator/{prompts/memGPT_function_schema.json => memgpt/prompts/function_schema.json} (98%) rename creator/{prompts/memGPT_prompt.md => memgpt/prompts/system_prompt.md} (100%) rename creator/{utils => memgpt}/time_utils.py (100%) create mode 100644 creator/memgpt/tools_handler.py diff --git a/creator/agents/__init__.py b/creator/agents/__init__.py index d185ef8..e4bce29 100644 --- a/creator/agents/__init__.py +++ b/creator/agents/__init__.py @@ -3,6 +3,7 @@ from .tester_agent import create_code_tester_agent from .refactor_agent import create_code_refactor_agent from .prompt_enhancer_agent import create_prompt_enhancer_agent +from .creator_agent import create_creator_agent __all__ = [ @@ -10,5 +11,6 @@ "create_code_interpreter_agent", "create_code_tester_agent", "create_code_refactor_agent", - "create_prompt_enhancer_agent" + "create_prompt_enhancer_agent", + "create_creator_agent" ] diff --git a/creator/agents/base.py b/creator/agents/base.py index 55a2abc..3aacbe5 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -2,7 +2,6 @@ from typing import List, Dict, Any, Optional from threading import Thread import json -import uuid from langchain.chains import LLMChain from langchain.callbacks.manager import CallbackManager @@ -14,7 +13,6 @@ from creator.utils import get_user_info, ask_run_code_confirm, remove_tips from creator.callbacks.buffer_manager import buffer_output_manager -from creator.config.library import config class BaseAgent(LLMChain): @@ -30,6 +28,9 @@ class BaseAgent(LLMChain): def _chain_type(self): return "BaseAgent" + def __repr__(self) -> str: + return self._chain_type + "()" + @property def input_keys(self) -> List[str]: return ["messages"] @@ -131,7 +132,7 @@ def run_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackMan langchain_messages.append(tool_result) langchain_messages = self.messages_hot_fix(langchain_messages) current_try += 1 - self.end_callbacks(message) + self.end_callbacks(message=message) langchain_messages = remove_tips(langchain_messages) openai_messages = list(map(convert_message_to_dict, langchain_messages)) return openai_messages diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py index c4210e1..6920155 100644 --- a/creator/agents/creator_agent.py +++ b/creator/agents/creator_agent.py @@ -10,7 +10,6 @@ from creator.code_interpreter.safe_python import SafePythonInterpreter from creator.config.library import config from creator.utils import load_system_prompt, get_user_info, remove_tips -from creator.llm import create_llm from .base import BaseAgent @@ -20,9 +19,8 @@ ALLOWED_FUNCTIONS = {"create", "save", "search", "CodeSkill"} ALLOW_METHODS = {".show", ".show_code", ".test", ".run", ".save", "__add__", "__gt__", "__lt__", "__annotations__"} IMPORT_CODE = ( - "from creator.core import creator\n" + "from creator import create, save, search\n" "from creator.core.skill import CodeSkill\n" - "create, save, search = creator.create, creator.save, creator.search\n\n" ) @@ -82,7 +80,3 @@ def create_creator_agent(llm): verbose=False, ) return chain - - -llm = create_llm(config) -open_creator_agent = create_creator_agent(llm=llm) diff --git a/creator/agents/memgpt_agent.py b/creator/agents/memgpt_agent.py deleted file mode 100644 index 3511451..0000000 --- a/creator/agents/memgpt_agent.py +++ /dev/null @@ -1,244 +0,0 @@ -from typing import List, Dict, Any, Optional -import datetime -import json -import uuid - -from langchain.schema.messages import FunctionMessage -from langchain.output_parsers.json import parse_partial_json -from langchain.callbacks.manager import CallbackManager -from langchain.prompts import ChatPromptTemplate -from langchain.adapters.openai import convert_openai_messages, convert_message_to_dict -from creator.config.library import config -from creator.utils import load_system_prompt, get_user_info -from creator.callbacks.streaming_stdout import OutputBufferStreamingHandler -from creator.agents.base import BaseAgent -from creator.utils.memgpt_utils import ( - get_initial_boot_messages, - get_login_event, - DEFAULT_PERSONA, - DEFAULT_HUMAN, - DEFAULT_AGENT_SUBTASKS, - package_message, - MESSAGE_SUMMARY_WARNING_TOKENS, - MESSAGE_SUMMARY_WARNING_STR -) -from creator.memory.schema import MessageConverter, ChatMessageHistory -from creator.llm.tokentrim import num_tokens_from_messages - - -class MemGPTAgent(BaseAgent): - total_tries: int = 10 - allow_user_confirm: bool = config.run_human_confirm - subagent: Optional[BaseAgent] = None - pause_heartbeats_start: Optional[datetime.datetime] = None - is_new_session: bool = True - heartbeat_request: bool = False - long_term_memory: ChatMessageHistory = None - - @property - def _chain_type(self): - return "MemGPTAgent" - - @property - def input_keys(self) -> List[str]: - return ["user_request"] - - def build_memory(self, session_id=None): - if session_id is None: - session_id = str(uuid.uuid4()) - self.long_term_memory = ChatMessageHistory( - session_id=session_id, - connection_string=f"sqlite:///{config.memory_path}/.langchain.db", - custom_message_converter=MessageConverter() - ) - - async def send_message(self, message: str, receiver: str = "human"): - """Sends a message to a specified receiver""" - if receiver == "human": - return None - else: - # run subagent - pass - - async def pause_heartbeats(self, minutes: int, max_pause: int = 360): - """Pauses heartbeats for a specified number of minutes""" - minutes = min(max_pause, minutes) - self.pause_heartbeats_start = datetime.datetime.now() - self.pause_heartbeats_minutes = int(minutes) - return f'Pausing timed heartbeats for {minutes} min' - - async def add_memory(self, name: str, content: str): - """Adds a memory with a specified name and content, and optionally requests a heartbeat""" - pass - - async def modify_memory(self, name: str, old_content:str, new_content: str): - """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" - pass - - async def search_memory(self, memory_type: str, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): - """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" - pass - - def get_callbacks(self): - callbacks = [] - if self.llm.callbacks is not None: - for callback in self.llm.callbacks.handlers: - if isinstance(callback, OutputBufferStreamingHandler): - callbacks.append(callback) - break - return callbacks - - def preprocess_inputs(self, inputs: Dict[str, Any]): - """Preprocesses inputs to the agent""" - session_id = inputs.get("session_id", None) - if session_id is None: - session_id = str(uuid.uuid4()) - inputs["session_id"] = session_id - self.is_new_session = True - if self.long_term_memory is None: - self.build_memory(session_id=session_id) - - inputs["memory_edit_timestamp"] = self.long_term_memory.memory_edit_timestamp - inputs["recall_memory_count"] = self.long_term_memory.recall_memory_count - inputs["archival_memory_count"] = self.long_term_memory.archival_memory_count - if inputs["recall_memory_count"] > 0: - self.is_new_session = False - - if "persona" not in inputs: - inputs["persona"] = DEFAULT_PERSONA - if "human" not in inputs: - inputs["human"] = DEFAULT_HUMAN - if "subagent_tasks" not in inputs: - inputs["subagent_tasks"] = DEFAULT_AGENT_SUBTASKS - - return inputs - - def construct_prompt(self, langchain_messages: Dict[str, Any]): - extra_messages = [] - if self.is_first: - init_messages = convert_openai_messages(get_initial_boot_messages()) - login_message = convert_openai_messages(get_login_event()) - extra_messages = init_messages + login_message - - prompt = ChatPromptTemplate.from_messages(messages=[ - ("system", self.system_template + get_user_info()), - *extra_messages, - *langchain_messages - ]) - return prompt - - - async def arun_tool(self, function_call: Dict[str, Any], run_manager: Optional[CallbackManager] = None): - function_name = function_call.get("name", "") - - available_functions = { - "pause_heartbeats": self.pause_heartbeats, - "send_message": self.send_message, - "add_memory": self.add_memory, - "modify_memory": self.modify_memory, - "search_memory": self.search_memory, - } - - # Failure case 1: function name is wrong - try: - function_to_call = available_functions[function_name] - except KeyError: - extra_info = {"status": 'Failed', "message": f'No function named {function_name}'} - function_response = package_message(message_type=None, extra_info=extra_info) - self.heartbeat_request = True - message = FunctionMessage(name=function_name, content=function_response) - return message - - # Failure case 2: function name is OK, but function args are bad JSON - try: - raw_function_args = function_call.get("arguments", "{}") - function_args = parse_partial_json(raw_function_args) - except Exception: - extra_info = {"status": 'Failed', "message": f"Error parsing JSON for function '{function_name}' arguments: {raw_function_args}"} - function_response = package_message(message_type=None, extra_info=extra_info) - self.heartbeat_request = True - message = FunctionMessage(name=function_name, content=function_response) - return message - - # (Still parsing function args) - # Handle requests for immediate heartbeat - heartbeat_request = function_args.pop('request_heartbeat', False) - if not (isinstance(heartbeat_request, bool) or heartbeat_request is None): - print(f"> Warning: 'request_heartbeat' arg parsed was not a bool or None, type={type(heartbeat_request)}, value={heartbeat_request}", print_type="markdown") - heartbeat_request = False - self.heartbeat_request = heartbeat_request - - # Failure case 3: function failed during execution - try: - function_response_string = await function_to_call(**function_args) - extra_info = {"status": 'OK', "message": function_response_string} - function_response = package_message(message_type=None, extra_info=extra_info) - except Exception as e: - extra_info = {"status": 'Failed', "message": f"Error calling function {function_name} with args {function_args}: {str(e)}"} - function_response = package_message(message_type=None, extra_info=extra_info) - self.heartbeat_request = True - message = FunctionMessage(name=function_name, content=function_response) - return message - - # If no failures happened along the way: ... - # Step 4: send the info on the function call and function response to GPT - message = FunctionMessage(name=function_name, content=function_response) - return message - - def check_token_limit(self, langchain_messages): - """Check if the token limit is reached, if so, return the messages with a warning message""" - token_cnt = num_tokens_from_messages(convert_message_to_dict(langchain_messages), model=config.model) - if token_cnt > self.token_limit: - warning_message = package_message(message_type="system_alert", extra_info={"message": MESSAGE_SUMMARY_WARNING_STR}) - langchain_messages.append(warning_message) - - async def arun_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: - run_manager_callbacks = run_manager.get_child() if run_manager else None - inputs = self.preprocess_inputs(inputs) - # handle usre request - user_request = inputs.get("user_request") - user_message = package_message(message_type="user_message", extra_info={"message": user_request}) - self.long_term_memory.add_user_message(user_message) - langchain_messages = self.long_term_memory.messages() - - # check token limit warning - # convert_message_to_dict(langchain_messages) - - # llm_with_functions = self.llm.bind(functions=self.function_schemas) - # current_try = 0 - # while current_try < self.total_tries: - # self.start_callbacks() - # prompt = self.construct_prompt(langchain_messages) - # llm_chain = (prompt | llm_with_functions | self.postprocess_mesasge).with_config({"run_name": f"Iteration {current_try+1}"}) - # message = llm_chain.invoke(inputs, {"callbacks": run_manager_callbacks}) - # langchain_messages.append(message) - # function_call = message.additional_kwargs.get("function_call", None) - # if function_call is None: - # self.end_callbacks(message) - # break - - # tool_result = self.run_tool(function_call, run_manager_callbacks) - # if tool_result is None: - # self.end_callbacks(message) - # break - # langchain_messages.append(tool_result) - # langchain_messages = self.messages_hot_fix(langchain_messages) - # current_try += 1 - # self.end_callbacks(message) - # langchain_messages = remove_tips(langchain_messages) - # openai_messages = list(map(convert_message_to_dict, langchain_messages)) - # return openai_messages - - -def create_memgpt_agent(llm, subagent=None): - template = load_system_prompt(config.memGPT_prompt_path) - with open(config.memGPT_schema_path) as f: - function_schemas = json.load(f) - chain = MemGPTAgent( - llm=llm, - subagent=subagent, - system_template=template, - function_schemas=function_schemas, - verbose=False, - ) - return chain diff --git a/creator/app/server.py b/creator/app/server.py index 70a0676..511a0e7 100644 --- a/creator/app/server.py +++ b/creator/app/server.py @@ -1,6 +1,7 @@ from fastapi import FastAPI -from creator.agents.creator_agent import create_llm, create_creator_agent -from creator.config.library import config +from creator.agents.creator_agent import create_creator_agent +from creator.llm import create_llm +from creator import config from creator.__version__ import __version__ as version from pydantic import BaseModel diff --git a/creator/app/streamlit_app.py b/creator/app/streamlit_app.py index 0d81bfe..bbb6870 100644 --- a/creator/app/streamlit_app.py +++ b/creator/app/streamlit_app.py @@ -4,8 +4,8 @@ sys.path.append(os.path.join(os.path.dirname(script_path), "../..")) import streamlit as st -from creator.agents.creator_agent import open_creator_agent -from creator.agents import code_interpreter_agent +from creator.agents import create_creator_agent, create_code_interpreter_agent +from creator.llm import create_llm from creator import config from langchain.callbacks.streamlit.streamlit_callback_handler import _convert_newlines from langchain.output_parsers.json import parse_partial_json @@ -21,6 +21,11 @@ agent_name = "interpreter_agent" +agent_mapping = { + "interpreter_agent": create_code_interpreter_agent(create_llm(config)), + "creator_agent": create_creator_agent(create_llm(config)) +} + def setup_slidebar(): global agent_name @@ -62,8 +67,6 @@ def setup_state(): if "langugae" not in st.session_state: st.session_state["language"] = "English" - if "agent" not in st.session_state: - st.session_state["agent"] = "interpreter_agent" def disable(): st.session_state["disabled"] = True @@ -173,7 +176,7 @@ def handle_input(): messages.append({"role": "user", "content": prompt}) print("current input messages", messages) render_conversation_history(container, messages) - agent = open_creator_agent if agent_name == "creator_agent" else code_interpreter_agent + agent = agent_mapping[agent_name] return_messages = stream_render(agent, messages, container) current_session["messages"] = return_messages diff --git a/creator/callbacks/streaming_stdout.py b/creator/callbacks/streaming_stdout.py index 0a457f6..bf8663b 100644 --- a/creator/callbacks/streaming_stdout.py +++ b/creator/callbacks/streaming_stdout.py @@ -1,4 +1,4 @@ -from typing import Any +from typing import Any, Dict, List from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler @@ -14,7 +14,7 @@ def on_chain_start(self, **kwargs: Any) -> None: agent_name = kwargs.get("agent_name") buffer_output_manager.add(agent_name) - def on_llm_new_token(self, **kwargs: Any) -> None: + def on_llm_new_token(self, token: str, **kwargs: Any) -> None: """Run on new LLM token. Only available when streaming is enabled.""" chunk = kwargs.get("chunk", None) if chunk is not None: @@ -41,7 +41,7 @@ def on_chain_start(self, **kwargs: Any) -> None: agent_name = kwargs.get("agent_name") rich_output_manager.add(agent_name) - def on_llm_new_token(self, **kwargs: Any) -> None: + def on_llm_new_token(self, token: str, **kwargs: Any) -> None: """Run on new LLM token. Only available when streaming is enabled.""" chunk = kwargs.get("chunk", None) if chunk is not None: @@ -68,7 +68,7 @@ def on_chain_start(self, **kwargs: Any) -> None: agent_name = kwargs.get("agent_name") file_output_manager.add(agent_name) - def on_llm_new_token(self, **kwargs: Any) -> None: + def on_llm_new_token(self, token: str, **kwargs: Any) -> None: """Run on new LLM token. Only available when streaming is enabled.""" chunk = kwargs.get("chunk", None) if chunk is not None: diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py index 1108de4..b6f9238 100644 --- a/creator/client/repl/handler.py +++ b/creator/client/repl/handler.py @@ -2,8 +2,7 @@ from .constants import help_commands, prompt_prefix, interpreter_prefix -from creator.agents.creator_agent import open_creator_agent -from creator.agents import create_code_interpreter_agent +from creator.agents import create_creator_agent, create_code_interpreter_agent from creator.config.library import config from creator.llm import create_llm from creator.utils import truncate_output, is_valid_code @@ -24,6 +23,7 @@ def __init__(self): self.console = Console() self.interpreter = False self.interpreter_agent = create_code_interpreter_agent(create_llm(config)) + self.open_creator_agent = create_creator_agent(create_llm(config)) async def handle(self, request, interpreter): """ @@ -37,8 +37,8 @@ async def handle(self, request, interpreter): """ self.interpreter = interpreter if request.startswith("%"): - await self.meta_prompt_handler(request, ) - elif is_valid_code(request, open_creator_agent.tools[0].namespace): + await self.meta_prompt_handler(request) + elif is_valid_code(request, self.open_creator_agent.tools[0].namespace): await self.expression_handler(request) self.update_history(request) else: @@ -128,7 +128,7 @@ async def expression_handler(self, request): str: The result of the executed expression or error message to be displayed in the output_field. """ - tool_result = open_creator_agent.tools[0].run_with_return(request) + tool_result = self.open_creator_agent.tools[0].run_with_return(request) truncate_tool_result = truncate_output(tool_result) outputs = [tool_result["stdout"], tool_result["stderr"]] output = "\n".join([o for o in outputs if o != ""]) @@ -194,7 +194,7 @@ async def agent_handler(self, request): if self.interpreter: messages = self.code_interpreter_agent.run(inputs) else: - messages = open_creator_agent.run(inputs) + messages = self.open_creator_agent.run(inputs) output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages)+1:]]) self.messages = messages self.output.append((output, "Markdown")) diff --git a/creator/config.yaml b/creator/config.yaml index 8d7f1c1..88482e5 100644 --- a/creator/config.yaml +++ b/creator/config.yaml @@ -4,7 +4,6 @@ LOCAL_SKILL_LIBRARY_VECTORD_PATH: .cache/open_creator/vectordb/ PROMPT_CACHE_HISTORY_PATH: .cache/open_creator/prompt_cache/ LOGGER_CACHE_PATH: .cache/open_creator/logs/ SKILL_EXTRACT_AGENT_CACHE_PATH: .cache/open_creator/llm_cache -MEMORY_PATH: .cache/open_creator/memory OFFICIAL_SKILL_LIBRARY_PATH: timedomain/skill-library OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH: timedomain/skill-library-template diff --git a/creator/config/library.py b/creator/config/library.py index 8e604da..9a69879 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -27,7 +27,6 @@ def resolve_path(path): _prompt_cache_history_path = resolve_path(yaml_config.get("PROMPT_CACHE_HISTORY_PATH", ".cache/open_creator/prompt_cache/")) _logger_cache_path = resolve_path(yaml_config.get("LOGGER_CACHE_PATH", ".cache/open_creator/logs/")) _skill_extract_agent_cache_path = resolve_path(yaml_config.get("SKILL_EXTRACT_AGENT_CACHE_PATH", ".cache/open_creator/llm_cache")) -_memory_path = resolve_path(yaml_config.get("MEMORY_PATH", ".cache/open_creator/memory")) _official_skill_library_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_PATH", "timedomain/skill-library")) _official_skill_library_template_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH", "timedomain/skill-library-template")) _model = yaml_config.get("MODEL_NAME", "gpt-3.5-turbo-16k-0613") @@ -38,7 +37,7 @@ def resolve_path(path): _build_in_skill_library_dir = os.path.join(project_dir, _build_in_skill_library_dir) # Ensure directories exist -for path in [_skill_extract_agent_cache_path, _local_skill_library_path, _local_skill_library_vectordb_path, _prompt_cache_history_path, _logger_cache_path, _memory_path]: +for path in [_skill_extract_agent_cache_path, _local_skill_library_path, _local_skill_library_vectordb_path, _prompt_cache_history_path, _logger_cache_path]: if not os.path.exists(path): os.makedirs(path) @@ -66,7 +65,6 @@ class LibraryConfig(BaseModel): prompt_cache_history_path: str = _prompt_cache_history_path logger_cache_path: str = _logger_cache_path skill_extract_agent_cache_path: str = _skill_extract_agent_cache_path - memory_path: str = _memory_path model: str = _model temperature: float = _temperature official_skill_library_path: str = _official_skill_library_path @@ -90,8 +88,6 @@ class LibraryConfig(BaseModel): tips_for_veryfy_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_veryfy_prompt.md") prompt_enhancer_agent_prompt_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_agent_prompt.md") prompt_enhancer_schema_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_schema.json") - memGPT_prompt_path: str = os.path.join(project_dir, "prompts", "memGPT_prompt.md") - memGPT_schema_path: str = os.path.join(project_dir, "prompts", "memGPT_function_schema.json") use_rich: bool = True use_file_logger: bool = False diff --git a/creator/llm/chatllm_with_trim.py b/creator/llm/chatllm_with_trim.py index 44086b2..94cbde9 100644 --- a/creator/llm/chatllm_with_trim.py +++ b/creator/llm/chatllm_with_trim.py @@ -5,21 +5,23 @@ class TrimMixin: - function_calls: Optional[List[Dict]] = None - def _create_message_dicts( self, messages: List[BaseMessage], stop: Optional[List[str]] ) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]: message_dicts, params = super()._create_message_dicts(messages, stop) - message_dicts = trim(messages=message_dicts, model=self.model_name, max_tokens=self.max_tokens, function_calls=self.function_calls) + message_dicts, self.trimed = trim(messages=message_dicts, model=self.model_name, max_tokens=self.max_tokens, function_calls=self.function_calls) return message_dicts, params class ChatOpenAIWithTrim(TrimMixin, ChatOpenAI): cache: bool = True + function_calls: Optional[List[Dict]] = None + trimed: bool = False pass class AzureChatOpenAIWithTrim(TrimMixin, AzureChatOpenAI): cache: bool = True + function_calls: Optional[List[Dict]] = None + trimed: bool = False pass diff --git a/creator/llm/format_function_calls.py b/creator/llm/format_function_calls.py index b014b6f..7f0906b 100644 --- a/creator/llm/format_function_calls.py +++ b/creator/llm/format_function_calls.py @@ -74,7 +74,7 @@ def format_schema(schema, indent, json_schema): elif schema["type"] == "array": return format_schema(schema["items"], indent, json_schema) + "[]" else: - raise ValueError("unknown schema type " + schema["type"]) + return schema["type"] def format_tool(tool): diff --git a/creator/llm/tokentrim.py b/creator/llm/tokentrim.py index 1199e99..08bafb1 100644 --- a/creator/llm/tokentrim.py +++ b/creator/llm/tokentrim.py @@ -145,8 +145,10 @@ def trim( """ Trim a list of messages to fit within a model's token limit. """ + trimed = False + if not messages: - return messages + return messages, trimed # Initialize max_tokens if max_tokens is None: @@ -158,14 +160,14 @@ def trim( total_tokens = num_tokens_from_messages(messages, model) if total_tokens <= max_tokens: - return messages + return messages, trimed # Deduct the system message tokens from the max_tokens if system message exists system_messages = [msg for msg in messages if msg["role"] == "system"] system_message_tokens = num_tokens_from_messages(system_messages, model) available_tokens = max_tokens - system_message_tokens - + trimed = True if available_tokens < 0: print("`tokentrim`: Warning, system message exceeds token limit. Trimming...") curr_tokens = total_tokens @@ -181,7 +183,7 @@ def trim( trimmed_messages.append(message) trimmed_messages.extend(messages[idx+1:]) break - return trimmed_messages + return trimmed_messages, trimed # trim except system messages idx = 0 @@ -201,4 +203,4 @@ def trim( idx += 1 break - return [msg for i, msg in enumerate(messages) if i not in removed_idxs] + return [msg for i, msg in enumerate(messages) if i not in removed_idxs], trimed diff --git a/creator/memgpt/__init__.py b/creator/memgpt/__init__.py new file mode 100644 index 0000000..729aa6a --- /dev/null +++ b/creator/memgpt/__init__.py @@ -0,0 +1,8 @@ +from .agent import create_memgpt +from .config import memory_config + + +__all__ = [ + "create_memgpt", + "memory_config" +] diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py new file mode 100644 index 0000000..276cb55 --- /dev/null +++ b/creator/memgpt/agent.py @@ -0,0 +1,230 @@ +from typing import List, Dict, Any, Optional +import datetime +import json + +from langchain.schema.messages import FunctionMessage +from langchain.output_parsers.json import parse_partial_json +from langchain.callbacks.manager import CallbackManager +from langchain.prompts import ChatPromptTemplate +from langchain.adapters.openai import convert_openai_messages + +from creator.utils import load_system_prompt +from creator.agents.base import BaseAgent + +from .memory import MemoryManager +from .constants import ( + DEFAULT_AGENT_SUBTASKS, + MESSAGE_SUMMARY_WARNING_STR, + FUNC_FAILED_HEARTBEAT_MESSAGE, + REQ_HEARTBEAT_MESSAGE, + MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE +) +from .message_handler import ( + get_initial_boot_messages, + get_login_event, + package_message +) +from .tools_handler import available_functions + + +class MemGPT(BaseAgent): + subagent: Optional[BaseAgent] = None + pause_heartbeats_start: Optional[datetime.datetime] = None + function_failed: bool = False + heartbeat_request: bool = False + memory_manager: MemoryManager = None + memgpt_config: dict = None + pause_heartbeats_start: datetime.datetime = None + pause_heartbeats_minutes: int = 0 + + @property + def _chain_type(self): + return "MemGPT" + + @property + def input_keys(self) -> List[str]: + return ["user_request"] + + def preprocess_inputs(self, inputs: Dict[str, Any]): + """Preprocesses inputs to the agent""" + if "session_id" in inputs and inputs["session_id"] != self.memgpt_config.session_id: + self.memgpt_config.session_id = inputs["session_id"] + self.memory_manager = MemoryManager(self.memgpt_config) + else: + inputs["session_id"] = self.memory_manager.session_id + + inputs["memory_edit_timestamp"] = self.memory_manager.memory_edit_timestamp + inputs["recall_memory_count"] = self.memory_manager.recall_memory_count + inputs["archival_memory_count"] = self.memory_manager.archival_memory_count + + inputs["persona"] = self.memory_manager.persona + inputs["human"] = self.memory_manager.human + if "subagent_tasks" not in inputs: + inputs["subagent_tasks"] = DEFAULT_AGENT_SUBTASKS + + return inputs + + def construct_prompt(self, langchain_messages: Dict[str, Any]): + extra_messages = [] + if self.memory_manager.is_new_session: + init_messages = convert_openai_messages(get_initial_boot_messages()) + login_message = convert_openai_messages(get_login_event()) + extra_messages = init_messages + login_message + + prompt = ChatPromptTemplate.from_messages(messages=[ + ("system", self.system_template), + *extra_messages, + *langchain_messages + ]) + return prompt + + async def arun_tool(self, function_call: Dict[str, Any], run_manager: Optional[CallbackManager] = None): + function_name = function_call.get("name", "") + # Failure case 1: function name is wrong + try: + function_to_call = available_functions[function_name] + except KeyError: + extra_info = {"status": 'Failed', "message": f'No function named {function_name}'} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = None + self.function_failed = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # Failure case 2: function name is OK, but function args are bad JSON + try: + raw_function_args = function_call.get("arguments", "{}") + function_args = parse_partial_json(raw_function_args) + except Exception: + extra_info = {"status": 'Failed', "message": f"Error parsing JSON for function '{function_name}' arguments: {raw_function_args}"} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = None + self.function_failed = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # (Still parsing function args) + # Handle requests for immediate heartbeat + heartbeat_request = function_args.pop('request_heartbeat', False) + if not (isinstance(heartbeat_request, bool) or heartbeat_request is None): + print(f"> Warning: 'request_heartbeat' arg parsed was not a bool or None, type={type(heartbeat_request)}, value={heartbeat_request}", print_type="markdown") + heartbeat_request = None + self.heartbeat_request = heartbeat_request + + # Failure case 3: function failed during execution + try: + function_args["memgpt"] = self + function_response_string = await function_to_call.arun(tool_input=function_args, callbacks=run_manager) + extra_info = {"status": 'OK', "message": function_response_string} + function_response = package_message(message_type=None, extra_info=extra_info) + except Exception as e: + extra_info = {"status": 'Failed', "message": f"Error calling function {function_name} with args {function_args}: {str(e)}"} + function_response = package_message(message_type=None, extra_info=extra_info) + self.heartbeat_request = None + self.function_failed = True + message = FunctionMessage(name=function_name, content=function_response) + return message + + # for send_message, when receiver is human, no need to send heartbeat request + if function_name == "send_message": + is_human = "receiver" not in function_args or function_args["receiver"] == "human" + if is_human: + self.heartbeat_request = None + + # If no failures happened along the way: ... + # Step 4: send the info on the function call and function response to GPT + message = FunctionMessage(name=function_name, content=function_response) + return message + + async def arun_workflow(self, inputs: Dict[str, Any], run_manager: Optional[CallbackManager] = None) -> Dict[str, Any]: + run_manager_callbacks = run_manager.get_child() if run_manager else None + self.llm.function_calls = self.function_schemas + llm_with_functions = self.llm.bind(functions=self.function_schemas) + user_request = inputs.get("user_request") + user_message = package_message(message_type="user_message", extra_info={"message": user_request}) + + counter = 0 + while True: + self.start_callbacks() + skil_next_user_input = False + + # we need re-preprocess inputs because core memory can be modified + inputs = self.preprocess_inputs(inputs) + + # handle usre request + await self.memory_manager.add_user_message(user_message) + langchain_messages = self.memory_manager.messages + + # construct prompt and run + prompt = self.construct_prompt(langchain_messages) + llm_chain = (prompt | llm_with_functions).with_config({"run_name": f"Iteration {counter+1}"}) + message = await llm_chain.ainvoke(inputs, {"callbacks": run_manager_callbacks}) + await self.memory_manager.add_message(message) + + # handle with ai response + function_call = message.additional_kwargs.get("function_call", None) + if function_call is None: + self.heartbeat_request = None + self.function_failed = None + else: + tool_result = await self.arun_tool(function_call, run_manager_callbacks) + await self.memory_manager.add_message(tool_result) + + user_message = None + if self.llm.trimed: + user_message = package_message(message_type="system_alert", extra_info={"message": MESSAGE_SUMMARY_WARNING_STR}) + skil_next_user_input = True + elif self.function_failed: + user_message = package_message(message_type="heartbeat", extra_info={"reason": FUNC_FAILED_HEARTBEAT_MESSAGE}) + skil_next_user_input = True + elif self.heartbeat_request: + user_message = package_message(message_type="heartbeat", extra_info={"reason": REQ_HEARTBEAT_MESSAGE}) + skil_next_user_input = True + + if not skil_next_user_input or user_message is None: + self.end_callbacks(message) + break + + if user_message is not None: + await self.memory_manager.add_user_message(user_message) + counter += 1 + self.end_callbacks(message) + + return self.memory_manager.session_id + + async def aparse_output(self, session_id): + return {self.output_key: session_id} + + async def _acall( + self, + inputs: Dict[str, Any], + run_manager: Optional[CallbackManager] = None, + ) -> Dict[str, str]: + output = {self.output_key: None} + try: + messages = await self.arun_workflow(inputs, run_manager) + output = await self.aparse_output(messages) + except Exception as e: + self.error_callbacks(e) + raise e + return output + + +def create_memgpt(llm, memgpt_config, subagent=None): + template = load_system_prompt(memgpt_config.system_prompt_path) + with open(memgpt_config.function_schema_path) as f: + function_schemas = json.load(f) + memory_manager = MemoryManager(memgpt_config) + if subagent is None: + subagent = BaseAgent(llm=llm, system_template=MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE) + chain = MemGPT( + llm=llm, + subagent=subagent, + system_template=template, + function_schemas=function_schemas, + memory_manager=memory_manager, + memgpt_config=memgpt_config, + output_key="session_id", + verbose=False, + ) + return chain diff --git a/creator/memgpt/config.py b/creator/memgpt/config.py new file mode 100644 index 0000000..ad5d71b --- /dev/null +++ b/creator/memgpt/config.py @@ -0,0 +1,39 @@ +import os +from .constants import ( + DEFAULT_PERSONA, + DEFAULT_HUMAN, + CORE_MEMORY_HUMAN_CHAR_LIMIT, + CORE_MEMORY_PERSONA_CHAR_LIMIT +) + + +class AttrDict(dict): + def __getattr__(self, key): + try: + return self[key] + except KeyError: + raise AttributeError(key) + + def __setattr__(self, key, value): + self[key] = value + + def __delattr__(self, key): + if key in self: + del self[key] + else: + raise AttributeError(key) + + +memory_config = AttrDict({ + "memory_path": os.path.expanduser("~/.cache/open_creator/memory"), + "system_prompt_path": os.path.join(os.path.dirname(__file__), "prompts", "system_prompt.md"), + "function_schema_path": os.path.join(os.path.dirname(__file__), "prompts", "function_schema.json"), + # replace with your own session_id + "session_id": None, + "persona": DEFAULT_PERSONA, + "human": DEFAULT_HUMAN, + "persona_char_limit": CORE_MEMORY_PERSONA_CHAR_LIMIT, + "human_char_limit": CORE_MEMORY_HUMAN_CHAR_LIMIT, + "page_size": 5, + "use_vector_search": True, +}) diff --git a/creator/memgpt/constants.py b/creator/memgpt/constants.py new file mode 100644 index 0000000..ffdd844 --- /dev/null +++ b/creator/memgpt/constants.py @@ -0,0 +1,38 @@ +DEFAULT_PERSONA = """ +The following is a blank slate starter persona, I need to expand this to develop my own personality. + +My name is MemGPT. +I am kind, thoughtful, and inquisitive. +""" + + +DEFAULT_HUMAN = """ +This is what I know so far about the user, I should expand this as I learn more about them. + +First name: Chad +Last name: ? +Gender: Male +Age: ? +Nationality: ? +Occupation: Computer science PhD student at UC Berkeley +Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO +""" + +DEFAULT_AGENT_SUBTASKS = """ + - create/save/search skill + - run/test/refactor skill + - show skill +""" + +# the number of tokens consumed in a call before a system warning goes to the agent +MESSAGE_SUMMARY_WARNING_TOKENS = 6000 +# Default memory limits +CORE_MEMORY_PERSONA_CHAR_LIMIT = 2000 +CORE_MEMORY_HUMAN_CHAR_LIMIT = 2000 + +MESSAGE_SUMMARY_WARNING_STR = "Warning: the conversation history will soon reach its maximum length and be trimmed. Make sure to save any important information from the conversation to your memory before it is removed." +FUNC_FAILED_HEARTBEAT_MESSAGE = "Function call failed" +REQ_HEARTBEAT_MESSAGE = "request_heartbeat == true" + +MESSAGE_CHATGPT_FUNCTION_MODEL = "gpt-3.5-turbo" +MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE = "You are a helpful assistant. Keep your responses short and concise." diff --git a/creator/memgpt/memory/__init__.py b/creator/memgpt/memory/__init__.py new file mode 100644 index 0000000..734c7b9 --- /dev/null +++ b/creator/memgpt/memory/__init__.py @@ -0,0 +1,78 @@ +from .schema import build_memory +from .core_memory import CoreMemory +from .recall_memory import RecallMemory +from .archival_memory import ArchivalMemory +from ..time_utils import get_local_time + + +class MemoryManager: + + def __init__(self, memory_config): + self.chat_message_history = build_memory(memory_path=memory_config.memory_path, session_id=memory_config.session_id) + self.session_id = self.chat_message_history.session_id + self.core_memory = CoreMemory( + persona=memory_config.persona, + human=memory_config.human, + persona_char_limit=memory_config.persona_char_limit, + human_char_limit=memory_config.human_char_limit, + ) + self.recall_memory = RecallMemory(message_database=self.chat_message_history, use_vector_search=memory_config.use_vector_search) + self.archival_memory = ArchivalMemory(message_database=self.chat_message_history, use_vector_search=memory_config.use_vector_search) + self.page_size = self.recall_memory.page_size = self.archival_memory.page_size = memory_config.page_size + + @property + def human(self): + return self.core_memory.human + + @property + def persona(self): + return self.core_memory.persona + + @property + def memory_edit_timestamp(self): + messages = self.chat_message_history.messages + now = get_local_time() + if messages: + return messages[-1].additional_kwargs.get("created_at", now) + return now + + @property + def recall_memory_count(self): + messages = self.chat_message_history.messages + return len([m for m in messages if m.type != "archival"]) + + @property + def archival_memory_count(self): + messages = self.chat_message_history.messages + return len([m for m in messages if m.type == "archival"]) + + @property + def is_new_session(self): + return len(self.chat_message_history.messages) == 0 + + @property + def messages(self): + messages = self.chat_message_history.messages + # filter archival and subagent + messages = [m for m in messages if m.type != "archival" and "subagent" not in m.additional_kwargs] + return messages + + async def add_user_message(self, message): + self.chat_message_history.add_user_message(message) + + async def add_message(self, message): + self.chat_message_history.add_message(message) + + def get_memory(self, memory_type): + memory = getattr(self, f"{memory_type}_memory", None) + assert memory is not None, f"Memory type {memory_type} not found" + return memory + + async def add(self, memory_type, message, name=None): + return await self.get_memory(memory_type=memory_type).add(message, name=name) + + async def modify(self, memory_type, old_content, new_content, name=None): + return await self.get_memory(memory_type=memory_type).modify(old_content=old_content, new_content=new_content, name=name) + + async def search(self, memory_type, query, page, start_date=None, end_date=None): + return await self.get_memory(memory_type=memory_type).search(query=query, page=page, start_date=start_date, end_date=end_date) diff --git a/creator/memgpt/memory/archival_memory.py b/creator/memgpt/memory/archival_memory.py new file mode 100644 index 0000000..0d3865b --- /dev/null +++ b/creator/memgpt/memory/archival_memory.py @@ -0,0 +1,20 @@ +from .schema import ArchivalMessage +from .recall_memory import RecallMemory + + +class ArchivalMemory(RecallMemory): + + def __repr__(self) -> str: + if len(self.message_database.messages) == 0: + memory_str = "" + else: + memory_str = "\n".join([d.content for d in self.message_database.messages]) + return f"\n### ARCHIVAL MEMORY ###\n{memory_str}" + + def _filter_messages(self): + """Utility to filter messages based on roles.""" + return [d for d in self.message_database.messages if d.type in ['archival']] + + async def add(self, message, name=None): + """Adds a new memory string. Optionally, a name can be provided.""" + self.message_database.add_message(ArchivalMessage(content=message)) diff --git a/creator/memory/base.py b/creator/memgpt/memory/base.py similarity index 94% rename from creator/memory/base.py rename to creator/memgpt/memory/base.py index a46303c..1a7f398 100644 --- a/creator/memory/base.py +++ b/creator/memgpt/memory/base.py @@ -12,8 +12,8 @@ def __repr__(self) -> str: pass @abstractmethod - async def add(self, memory_string, name=None): - """Adds a new memory string. Optionally, a name can be provided.""" + async def add(self, message, name=None): + """Adds a new message. Optionally, a name can be provided.""" pass @abstractmethod diff --git a/creator/memory/short_term.py b/creator/memgpt/memory/core_memory.py similarity index 94% rename from creator/memory/short_term.py rename to creator/memgpt/memory/core_memory.py index 04388cf..e1bc1fc 100644 --- a/creator/memory/short_term.py +++ b/creator/memgpt/memory/core_memory.py @@ -36,8 +36,8 @@ def _edit(self, memory_string, name): setattr(self, name, memory_string) return len(memory_string) - async def add(self, memory_string, name): - new_content = getattr(self, name) + "\n" + memory_string + async def add(self, message, name): + new_content = getattr(self, name) + "\n" + message return self._edit(new_content, name) async def modify(self, old_content, new_content, name): diff --git a/creator/memory/long_term.py b/creator/memgpt/memory/recall_memory.py similarity index 69% rename from creator/memory/long_term.py rename to creator/memgpt/memory/recall_memory.py index b73d028..395e174 100644 --- a/creator/memory/long_term.py +++ b/creator/memgpt/memory/recall_memory.py @@ -1,12 +1,11 @@ import numpy as np from creator.llm import create_embedding +from creator.config.library import config from creator.retrivever.score_functions import cosine_similarity - from langchain.memory.chat_message_histories import SQLChatMessageHistory from .base import BaseMemory -from .schema import ArchivalMessage class RecallMemory(BaseMemory): @@ -19,13 +18,12 @@ class RecallMemory(BaseMemory): effectively allowing it to 'remember' prior engagements with a user. """ - def __init__(self, message_database: SQLChatMessageHistory, use_vector_search=False, page_size=5): + def __init__(self, message_database: SQLChatMessageHistory, use_vector_search=False): self.message_database = message_database self.use_vector_search = use_vector_search if use_vector_search: self.embeddings = dict() - self.embedding_model = create_embedding() - self.page_size = page_size + self.embedding_model = create_embedding(config) def __len__(self): return len(self.message_database.messages) @@ -47,10 +45,10 @@ def __repr__(self) -> str: return f"\n### RECALL MEMORY ###\nStatistics:\n{len(self.message_database.messages)} total messages\n{memory_str}" async def add(self, message, name=None): - raise NotImplementedError("Recall memory is read-only, will automatically update, and doesn't support editting or adding new memories as it's the record of your past interactions.") + self.message_database.add_message(message) async def modify(self, old_content, new_content, name=None): - raise NotImplementedError("Recall memory is read-only, will automatically update, and doesn't support editting or adding new memories as it's the record of your past interactions.") + raise NotImplementedError("Archival/Recall memory doesn't support modify!") def _filter_messages(self): """Utility to filter messages based on roles.""" @@ -80,24 +78,3 @@ async def search(self, query, page, start_date=None, end_date=None): else: matches = [d for d in matches if d.content and query.lower() in d.content.lower()] return self._paginate_results(matches, page) - - -class ArchivalMemory(RecallMemory): - - def __repr__(self) -> str: - if len(self.message_database.messages) == 0: - memory_str = "" - else: - memory_str = "\n".join([d.content for d in self.message_database.messages]) - return f"\n### ARCHIVAL MEMORY ###\n{memory_str}" - - def _filter_messages(self): - """Utility to filter messages based on roles.""" - return [d for d in self.message_database.messages if d.type in ['archival']] - - async def add(self, message: str, name=None): - """Adds a new memory string. Optionally, a name can be provided.""" - return self.message_database.add_message(ArchivalMessage(content=message)) - - async def modify(self, old_content, new_content, name=None): - raise NotImplementedError("Archival memory doesn't support modify!") diff --git a/creator/memory/schema.py b/creator/memgpt/memory/schema.py similarity index 68% rename from creator/memory/schema.py rename to creator/memgpt/memory/schema.py index b25ff1f..7791615 100644 --- a/creator/memory/schema.py +++ b/creator/memgpt/memory/schema.py @@ -1,5 +1,6 @@ from typing import Any import datetime +import uuid from sqlalchemy.orm import declarative_base from sqlalchemy import Column, Integer, Text, DateTime, JSON @@ -7,7 +8,9 @@ from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, FunctionMessage from langchain.memory.chat_message_histories.sql import BaseMessageConverter from langchain.memory.chat_message_histories import SQLChatMessageHistory -from creator.utils.time_utils import get_local_time + +from ..time_utils import get_local_time + Base = declarative_base() @@ -30,36 +33,40 @@ class MemoryMessage(Base): class MessageConverter(BaseMessageConverter): def from_sql_model(self, sql_message: Any) -> BaseMessage: - created_at = sql_message.additional_kwargs.get("created_at", datetime.now()) + created_at = sql_message.created_at.strftime("%Y-%m-%d %I:%M:%S %p %Z%z") sql_message.additional_kwargs.update({"created_at": created_at}) if sql_message.type == 'human': return HumanMessage( content=sql_message.content, - additional_kwargs={"created_at": sql_message.created_at} + additional_kwargs=sql_message.additional_kwargs ) - elif sql_message.type == 'ai': + elif "ai" in sql_message.type.lower(): return AIMessage( content=sql_message.content, additional_kwargs=sql_message.additional_kwargs ) - elif sql_message.type == 'system': + elif "system" in sql_message.type.lower(): return SystemMessage( content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs ) - elif sql_message.type == 'function': + elif "function" in sql_message.type.lower(): return FunctionMessage( content=sql_message.content, name=sql_message.additional_kwargs.get("name", ""), + additional_kwargs=sql_message.additional_kwargs ) - elif sql_message.type == "archival": + elif "archival" in sql_message.type.lower(): return ArchivalMessage( content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs ) else: raise ValueError(f'Unknown message type: {sql_message.type}') def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: now = get_local_time() + now_datetime = datetime.datetime.strptime(now, "%Y-%m-%d %I:%M:%S %p %Z%z") if isinstance(message, FunctionMessage): message.additional_kwargs.update({"name": message.name}) message.additional_kwargs.update({"created_at": now}) @@ -67,7 +74,7 @@ def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: session_id=session_id, type=message.type, content=message.content, - created_at=message.additional_kwargs.get("created_at", now), + created_at=now_datetime, additional_kwargs=message.additional_kwargs ) @@ -75,22 +82,11 @@ def get_sql_model_class(self) -> Any: return MemoryMessage -class ChatMessageHistory(SQLChatMessageHistory): - - @property - def memory_edit_timestamp(self): - messages = self.messages() - now = get_local_time() - if messages: - return messages[-1].additional_kwargs.get("created_at", now) - return now - - @property - def recall_memory_count(self): - messages = self.messages() - return len([m for m in messages if m.type != "archival"]) - - @property - def archival_memory_count(self): - messages = self.messages() - return len([m for m in messages if m.type == "archival"]) +def build_memory(memory_path:str, session_id=None): + if session_id is None: + session_id = str(uuid.uuid4()) + return SQLChatMessageHistory( + session_id=session_id, + connection_string=f"sqlite:///{memory_path}/.langchain.db", + custom_message_converter=MessageConverter() + ) diff --git a/creator/memgpt/message_handler.py b/creator/memgpt/message_handler.py new file mode 100644 index 0000000..6102294 --- /dev/null +++ b/creator/memgpt/message_handler.py @@ -0,0 +1,52 @@ +import json +from .time_utils import get_local_time + + +def package_message(message_type=None, extra_info=None, include_location=False, location_name='San Francisco, CA, USA', timestamp=None): + """ + A generic function to package different types of messages. + + :param message_type: The type of the message (e.g. 'heartbeat', 'login', 'user_message', 'system_alert'). + :param extra_info: A dictionary containing additional information specific to the message type. + :param include_location: A boolean indicating whether to include location information. + :param location_name: The name of the location to include if include_location is True. + :param timestamp: The timestamp for the message. If None, the current local time is used. + """ + + formatted_time = get_local_time() if timestamp is None else timestamp + # packaged_message = {"time": formatted_time} + packaged_message = {} + if message_type is not None: + packaged_message["type"] = message_type + + if include_location: + packaged_message["location"] = location_name + + if extra_info: + packaged_message.update(extra_info) + + return json.dumps(packaged_message) + + +def get_initial_boot_messages(): + return [ + # first message includes both inner monologue and function call to send_message + { + "role": "assistant", + "content": "Bootup sequence complete. Persona activated. Testing messaging functionality.", + "function_call": { + "name": "send_message", + "arguments": "{\n \"message\": \"More human than human is our motto.\"\n}" + } + }, + # obligatory function return message + { + "role": "function", + "name": "send_message", + "content": package_message(None, {"message": None, "status": "OK"}) + } + ] + + +def get_login_event(): + return package_message('login', {"last_login": 'Never (first login)'}, include_location=True) diff --git a/creator/prompts/memGPT_function_schema.json b/creator/memgpt/prompts/function_schema.json similarity index 98% rename from creator/prompts/memGPT_function_schema.json rename to creator/memgpt/prompts/function_schema.json index a9cbc12..01d9a67 100644 --- a/creator/prompts/memGPT_function_schema.json +++ b/creator/memgpt/prompts/function_schema.json @@ -7,7 +7,7 @@ "properties": { "message": { "type": "string", - "description": "Message contents. All unicode (including emojis) are supported." + "description": "Message contents. All unicode (including emojis) are supported. Reply in the user's language." }, "receiver": { "type": "string", diff --git a/creator/prompts/memGPT_prompt.md b/creator/memgpt/prompts/system_prompt.md similarity index 100% rename from creator/prompts/memGPT_prompt.md rename to creator/memgpt/prompts/system_prompt.md diff --git a/creator/utils/time_utils.py b/creator/memgpt/time_utils.py similarity index 100% rename from creator/utils/time_utils.py rename to creator/memgpt/time_utils.py diff --git a/creator/memgpt/tools_handler.py b/creator/memgpt/tools_handler.py new file mode 100644 index 0000000..e381a99 --- /dev/null +++ b/creator/memgpt/tools_handler.py @@ -0,0 +1,76 @@ +from typing import Union, Dict, Any +import math +import json + +from langchain.adapters.openai import convert_dict_to_message +from langchain.tools import StructuredTool + +from .time_utils import get_local_time + + +class Tool(StructuredTool): + + def _parse_input( + self, + tool_input: Union[str, Dict], + ) -> Union[str, Dict[str, Any]]: + """Return the original tool_input""" + return tool_input + + +async def send_message(memgpt, message: str, receiver: str = "human"): + """Sends a message to a specified receiver""" + if receiver == "human": + return + else: + # request subagent first + messages = memgpt.subagent.run({"messages": {"role": "user", "content": message}}) + for m in messages: + langchain_message = convert_dict_to_message(m) + langchain_message.additional_kwargs["subagent"] = memgpt.subagent._chain_type + memgpt.memory_manager.add_message(message) + last_m = messages[-1] + return last_m.content + + +async def pause_heartbeats(memgpt, minutes: int, max_pause: int = 360): + """Pauses heartbeats for a specified number of minutes""" + minutes = min(max_pause, minutes) + memgpt.pause_heartbeats_start = get_local_time() + memgpt.pause_heartbeats_minutes = int(minutes) + return f'Pausing timed heartbeats for {minutes} min' + + +async def add_memory(memgpt, name: str, content: str): + """Adds a memory with a specified name and content, and optionally requests a heartbeat""" + if name == "archival": + memory_type = "archival" + else: + memory_type = "core" + await memgpt.memory_manager.add(memory_type, content, name) + + +async def modify_memory(memgpt, name: str, old_content:str, new_content: str): + """Modifies a memory with a specified name, replacing old content with new content, and optionally requests a heartbeat""" + # only core memory can be modified + memory_type = "core" + await memgpt.memory_manager.modify(memory_type, old_content, new_content, name) + + +async def search_memory(memgpt, memory_type: str, page: int = 0, query: str = "", start_date: str = "", end_date: str = ""): + """Searches memory of a specified type, with optional query, start date, end date, and request for heartbeat""" + memory_type = "recall" if memory_type == "conversation" else memory_type + results, total = await memgpt.memory_manager.search(memory_type, query, page, start_date, end_date) + results_str = "" + if len(results) == 0: + results_str = "No results found" + else: + num_pages = math.ceil(total / memgpt.memory_manager.page_size) - 1 # 0 index + results_pref = f"Showing {len(results)} of {total} results (page {page}/{num_pages}):" + results_formatted = [f"timestamp: {d['timestamp']}, memory: {d['content']}" for d in results] + results_str = f"{results_pref} {json.dumps(results_formatted)}" + return results_str + + +tools = [Tool.from_function(coroutine=func) for func in [send_message, pause_heartbeats, add_memory, modify_memory, search_memory]] +available_functions = {tool.name:tool for tool in tools} diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index 5e7a0d2..7d89f80 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -11,7 +11,6 @@ from .code_split import split_code_blocks from .valid_code import is_valid_code, is_expression from .tips_utils import remove_tips -from .time_utils import get_local_time from .runnable_decorator import runnable, print_run_url @@ -30,7 +29,6 @@ "is_valid_code", "is_expression", "remove_tips", - "get_local_time", "runnable", "print_run_url" ] From 48d13ee1b853ec02a8e20c725068c4d03018a73c Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Mon, 30 Oct 2023 19:02:30 +0800 Subject: [PATCH 39/56] feat: support vector db qdrant --- creator/config.yaml | 37 ++++- creator/config/library.py | 69 +++++---- creator/config/load_config.py | 7 +- creator/core/core.py | 4 +- creator/llm/__init__.py | 5 +- creator/llm/llm_creator.py | 19 --- creator/memgpt/__init__.py | 4 +- creator/memgpt/config.py | 39 ----- creator/memgpt/constants.py | 27 ---- creator/memgpt/memory/__init__.py | 2 +- creator/memgpt/memory/archival_memory.py | 2 +- creator/memgpt/memory/builder.py | 15 ++ creator/memgpt/memory/converter.py | 64 ++++++++ creator/memgpt/memory/recall_memory.py | 31 ++-- creator/memgpt/memory/schema.py | 88 ++--------- .../memgpt_function_schema.json} | 0 .../memgpt_system_prompt.md} | 0 creator/retrivever/base.py | 137 +++++------------- creator/retrivever/embedding_creator.py | 20 +++ creator/retrivever/memory_retrivever.py | 21 +++ creator/retrivever/score_functions.py | 7 - creator/retrivever/skill_retrivever.py | 39 +++++ .../create/conversation_history.json | 6 - .../open-creator/create/embedding_text.txt | 22 --- .../open-creator/create/function_call.json | 46 ------ .../create/install_dependencies.sh | 1 - .../open-creator/create/skill.json | 106 -------------- .../open-creator/create/skill_code.py | 61 -------- .../open-creator/create/skill_doc.md | 26 ---- .../save/conversation_history.json | 6 - .../open-creator/save/embedding_text.txt | 5 - .../open-creator/save/function_call.json | 24 --- .../open-creator/save/install_dependencies.sh | 1 - .../open-creator/save/skill.json | 59 -------- .../open-creator/save/skill_code.py | 30 ---- .../open-creator/save/skill_doc.md | 26 ---- .../search/conversation_history.json | 6 - .../open-creator/search/embedding_text.txt | 4 - .../open-creator/search/function_call.json | 26 ---- .../search/install_dependencies.sh | 1 - .../open-creator/search/search.py | 22 --- .../open-creator/search/skill.json | 64 -------- .../open-creator/search/skill_code.py | 20 --- .../open-creator/search/skill_doc.md | 18 --- creator/utils/__init__.py | 6 +- creator/utils/attri_dict.py | 21 +++ creator/utils/uuid_generator.py | 23 +++ pyproject.toml | 2 +- 48 files changed, 357 insertions(+), 912 deletions(-) delete mode 100644 creator/memgpt/config.py create mode 100644 creator/memgpt/memory/builder.py create mode 100644 creator/memgpt/memory/converter.py rename creator/{memgpt/prompts/function_schema.json => prompts/memgpt_function_schema.json} (100%) rename creator/{memgpt/prompts/system_prompt.md => prompts/memgpt_system_prompt.md} (100%) create mode 100644 creator/retrivever/embedding_creator.py create mode 100644 creator/retrivever/memory_retrivever.py delete mode 100644 creator/retrivever/score_functions.py create mode 100644 creator/retrivever/skill_retrivever.py delete mode 100644 creator/skill_library/open-creator/create/conversation_history.json delete mode 100644 creator/skill_library/open-creator/create/embedding_text.txt delete mode 100644 creator/skill_library/open-creator/create/function_call.json delete mode 100644 creator/skill_library/open-creator/create/install_dependencies.sh delete mode 100644 creator/skill_library/open-creator/create/skill.json delete mode 100644 creator/skill_library/open-creator/create/skill_code.py delete mode 100644 creator/skill_library/open-creator/create/skill_doc.md delete mode 100644 creator/skill_library/open-creator/save/conversation_history.json delete mode 100644 creator/skill_library/open-creator/save/embedding_text.txt delete mode 100644 creator/skill_library/open-creator/save/function_call.json delete mode 100644 creator/skill_library/open-creator/save/install_dependencies.sh delete mode 100644 creator/skill_library/open-creator/save/skill.json delete mode 100644 creator/skill_library/open-creator/save/skill_code.py delete mode 100644 creator/skill_library/open-creator/save/skill_doc.md delete mode 100644 creator/skill_library/open-creator/search/conversation_history.json delete mode 100644 creator/skill_library/open-creator/search/embedding_text.txt delete mode 100644 creator/skill_library/open-creator/search/function_call.json delete mode 100644 creator/skill_library/open-creator/search/install_dependencies.sh delete mode 100644 creator/skill_library/open-creator/search/search.py delete mode 100644 creator/skill_library/open-creator/search/skill.json delete mode 100644 creator/skill_library/open-creator/search/skill_code.py delete mode 100644 creator/skill_library/open-creator/search/skill_doc.md create mode 100644 creator/utils/attri_dict.py create mode 100644 creator/utils/uuid_generator.py diff --git a/creator/config.yaml b/creator/config.yaml index 88482e5..eb4e8e5 100644 --- a/creator/config.yaml +++ b/creator/config.yaml @@ -1,13 +1,13 @@ LOCAL_SKILL_LIBRARY_PATH: .cache/open_creator/skill_library REMOTE_SKILL_LIBRARY_PATH: .cache/open_creator/remote -LOCAL_SKILL_LIBRARY_VECTORD_PATH: .cache/open_creator/vectordb/ PROMPT_CACHE_HISTORY_PATH: .cache/open_creator/prompt_cache/ +VECTORD_PATH: .cache/open_creator/vectordb/ LOGGER_CACHE_PATH: .cache/open_creator/logs/ -SKILL_EXTRACT_AGENT_CACHE_PATH: .cache/open_creator/llm_cache +LLM_CACHE_PATH: .cache/open_creator/llm_cache +EMBEDDING_CACHE_PATH: .cache/open_creator/embeddings/ OFFICIAL_SKILL_LIBRARY_PATH: timedomain/skill-library OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH: timedomain/skill-library-template -BUILD_IN_SKILL_LIBRARY_DIR: skill_library/open-creator/ # for AZURE, it is your_deployment_id # for ANTHROPIC, it is claude-2 @@ -39,3 +39,34 @@ LANGCHAIN_ENDPOINT: LANGCHAIN_API_KEY: LANGCHAIN_TRACING_V2: true LANGCHAIN_PROJECT: "open-creator" + +# for memgpt +MEMGPT_CONFIG: + MEMORY_PATH: .cache/open_creator/memory + PERSONA: | + The following is a blank slate starter persona, I need to expand this to develop my own personality. + + My name is MemGPT. + I am kind, thoughtful, and inquisitive. + + HUMAN: | + This is what I know so far about the user, I should expand this as I learn more about them. + + First name: Chad + Last name: ? + Gender: Male + Age: ? + Nationality: ? + Occupation: Computer science PhD student at UC Berkeley + Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO + + AGENT_SUBTASKS: | + - create/save/search skill + - run/test/refactor skill + - show skill + + SUMMARY_WARNING_TOKENS: 6000 + CORE_MEMORY_PERSONA_CHAR_LIMIT: 2000 + CORE_MEMORY_HUMAN_CHAR_LIMIT: 2000 + PAGE_SIZE: 5 + USE_VECTOR_SEARCH: true diff --git a/creator/config/library.py b/creator/config/library.py index 9a69879..ee88a10 100644 --- a/creator/config/library.py +++ b/creator/config/library.py @@ -1,9 +1,11 @@ +import os +from pydantic import BaseModel + from langchain.cache import SQLiteCache from langchain.globals import set_llm_cache -from pydantic import BaseModel + from creator.code_interpreter import CodeInterpreter from creator.config.load_config import load_yaml_config -import os # Load configuration from YAML @@ -21,23 +23,27 @@ def resolve_path(path): # Fetch values from the loaded YAML config or set default values -_local_skill_library_path = resolve_path(yaml_config.get("LOCAL_SKILL_LIBRARY_PATH", ".cache/open_creator/skill_library")) -_remote_skill_library_path = resolve_path(yaml_config.get("REMOTE_SKILL_LIBRARY_PATH", ".cache/open_creator/remote")) -_local_skill_library_vectordb_path = resolve_path(yaml_config.get("LOCAL_SKILL_LIBRARY_VECTORD_PATH", ".cache/open_creator/vectordb/")) -_prompt_cache_history_path = resolve_path(yaml_config.get("PROMPT_CACHE_HISTORY_PATH", ".cache/open_creator/prompt_cache/")) -_logger_cache_path = resolve_path(yaml_config.get("LOGGER_CACHE_PATH", ".cache/open_creator/logs/")) -_skill_extract_agent_cache_path = resolve_path(yaml_config.get("SKILL_EXTRACT_AGENT_CACHE_PATH", ".cache/open_creator/llm_cache")) -_official_skill_library_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_PATH", "timedomain/skill-library")) -_official_skill_library_template_path = resolve_path(yaml_config.get("OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH", "timedomain/skill-library-template")) -_model = yaml_config.get("MODEL_NAME", "gpt-3.5-turbo-16k-0613") -_temperature = yaml_config.get("TEMPERATURE", 0) -_run_human_confirm = yaml_config.get("RUN_HUMAN_CONFIRM", False) -_use_stream_callback = yaml_config.get("USE_STREAM_CALLBACK", True) -_build_in_skill_library_dir = yaml_config.get("BUILD_IN_SKILL_LIBRARY_DIR", "skill_library/open-creator/") -_build_in_skill_library_dir = os.path.join(project_dir, _build_in_skill_library_dir) +_local_skill_library_path = resolve_path(yaml_config.LOCAL_SKILL_LIBRARY_PATH) +_remote_skill_library_path = resolve_path(yaml_config.REMOTE_SKILL_LIBRARY_PATH) +_vectordb_path = resolve_path(yaml_config.VECTORD_PATH) +_prompt_cache_history_path = resolve_path(yaml_config.PROMPT_CACHE_HISTORY_PATH) +_logger_cache_path = resolve_path(yaml_config.LOGGER_CACHE_PATH) +_llm_cache_path = resolve_path(yaml_config.LLM_CACHE_PATH) +_embedding_cache_path = resolve_path(yaml_config.EMBEDDING_CACHE_PATH) +yaml_config.MEMGPT_CONFIG.MEMORY_PATH = resolve_path(yaml_config.MEMGPT_CONFIG.MEMORY_PATH) + +memgpt_config = yaml_config.MEMGPT_CONFIG + +_official_skill_library_path = resolve_path(yaml_config.OFFICIAL_SKILL_LIBRARY_PATH) +_official_skill_library_template_path = resolve_path(yaml_config.OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH) + +_model = yaml_config.MODEL_NAME +_temperature = yaml_config.TEMPERATURE +_run_human_confirm = yaml_config.RUN_HUMAN_CONFIRM +_use_stream_callback = yaml_config.USE_STREAM_CALLBACK # Ensure directories exist -for path in [_skill_extract_agent_cache_path, _local_skill_library_path, _local_skill_library_vectordb_path, _prompt_cache_history_path, _logger_cache_path]: +for path in [_llm_cache_path, _local_skill_library_path, _vectordb_path, _prompt_cache_history_path, _logger_cache_path, memgpt_config.MEMORY_PATH, _embedding_cache_path]: if not os.path.exists(path): os.makedirs(path) @@ -48,30 +54,24 @@ def resolve_path(path): if not os.path.exists(_prompt_cache_history_path): open(os.path.join(_prompt_cache_history_path, "history.txt"), 'a').close() -build_in_skill_library_dir = os.path.join(os.path.dirname(os.path.abspath(__file__))) - -build_in_skill_config = { - "create": os.path.join(_build_in_skill_library_dir, "create"), - "save": os.path.join(_build_in_skill_library_dir, "save"), - "search": os.path.join(_build_in_skill_library_dir, "search"), - -} # Placeholder for any built-in skill configurations - class LibraryConfig(BaseModel): local_skill_library_path: str = _local_skill_library_path remote_skill_library_path: str = _remote_skill_library_path - local_skill_library_vectordb_path: str = _local_skill_library_vectordb_path + vectordb_path: str = _vectordb_path prompt_cache_history_path: str = _prompt_cache_history_path logger_cache_path: str = _logger_cache_path - skill_extract_agent_cache_path: str = _skill_extract_agent_cache_path + llm_cache_path: str = _llm_cache_path + embedding_cache_path: str = _embedding_cache_path + model: str = _model temperature: float = _temperature - official_skill_library_path: str = _official_skill_library_path - official_skill_library_template_path: str = _official_skill_library_template_path - build_in_skill_config: dict = build_in_skill_config run_human_confirm: bool = _run_human_confirm use_stream_callback: bool = _use_stream_callback + + official_skill_library_path: str = _official_skill_library_path + official_skill_library_template_path: str = _official_skill_library_template_path + code_interpreter: CodeInterpreter = CodeInterpreter() # prompt paths @@ -88,11 +88,16 @@ class LibraryConfig(BaseModel): tips_for_veryfy_prompt_path: str = os.path.join(project_dir, "prompts", "tips_for_veryfy_prompt.md") prompt_enhancer_agent_prompt_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_agent_prompt.md") prompt_enhancer_schema_path: str = os.path.join(project_dir, "prompts", "prompt_enhancer_schema.json") + memgpt_system_prompt_path: str = os.path.join(project_dir, "prompts", "memgpt_system_prompt.md") + memgpt_function_schema_path: str = os.path.join(project_dir, "prompts", "memgpt_function_schema.json") + + memgpt_config: dict = None use_rich: bool = True use_file_logger: bool = False config = LibraryConfig() +config.memgpt_config = memgpt_config -set_llm_cache(SQLiteCache(database_path=f"{config.skill_extract_agent_cache_path}/.langchain.db")) +set_llm_cache(SQLiteCache(database_path=f"{config.llm_cache_path}/.langchain.db")) diff --git a/creator/config/load_config.py b/creator/config/load_config.py index 8ea20af..068235a 100644 --- a/creator/config/load_config.py +++ b/creator/config/load_config.py @@ -2,8 +2,11 @@ import shutil import appdirs import yaml + from dotenv import load_dotenv, find_dotenv +from ..utils.attri_dict import AttrDict + load_dotenv(find_dotenv()) @@ -41,6 +44,8 @@ def load_yaml_config(): with open(user_config_path, mode='r', encoding="utf-8") as f: user_config = yaml.safe_load(f) + if user_config is None: + user_config = {} updated = False for key, value in project_config.items(): @@ -61,4 +66,4 @@ def load_yaml_config(): if not os.environ.get(key) and value: os.environ[key] = str(value) - return user_config + return AttrDict(user_config) diff --git a/creator/core/core.py b/creator/core/core.py index 550a3ad..ee87281 100644 --- a/creator/core/core.py +++ b/creator/core/core.py @@ -3,7 +3,7 @@ from creator.config.library import config from creator.utils import print from creator.hub.huggingface import hf_pull -from creator.retrivever.base import BaseVectorStore +from creator.retrivever.skill_retrivever import SkillVectorStore from .skill import CodeSkill, BaseSkill, BaseSkillMetadata from .runnable import create_skill_from_messages, create_skill_from_request, create_skill_from_file_content @@ -184,7 +184,7 @@ def search(self, query: str, top_k: int = 3, threshold=0.8, remote=False) -> Lis raise NotImplementedError if self.vectordb is None: print("> loading vector database...", print_type="markdown") - self.vectordb = BaseVectorStore() + self.vectordb = SkillVectorStore() skills = self.vectordb.search(query, top_k=top_k, threshold=threshold) return [CodeSkill(**skill) if skill.get("skill_program_language", None) else BaseSkill(**skill) for skill in skills] diff --git a/creator/llm/__init__.py b/creator/llm/__init__.py index e28e5ec..ac8fd60 100644 --- a/creator/llm/__init__.py +++ b/creator/llm/__init__.py @@ -1,7 +1,6 @@ -from .llm_creator import create_llm, create_embedding +from .llm_creator import create_llm __all__ = [ - "create_llm", - "create_embedding" + "create_llm" ] diff --git a/creator/llm/llm_creator.py b/creator/llm/llm_creator.py index 59c6a37..3de5513 100644 --- a/creator/llm/llm_creator.py +++ b/creator/llm/llm_creator.py @@ -1,8 +1,6 @@ import os from creator.callbacks import OutputBufferStreamingHandler, RichTerminalStreamingHandler, FileLoggerStreamingHandler from langchain.callbacks.manager import CallbackManager -from langchain.embeddings import OpenAIEmbeddings, CacheBackedEmbeddings -from langchain.storage import LocalFileStore from .chatllm_with_trim import ChatOpenAIWithTrim, AzureChatOpenAIWithTrim @@ -32,20 +30,3 @@ def create_llm(config): streaming=streaming ) return llm - - -def create_embedding(config): - - use_azure = True if os.getenv("OPENAI_API_TYPE", None) == "azure" else False - - if use_azure: - azure_model = os.getenv("EMBEDDING_DEPLOYMENT_NAME", None) - print(azure_model) - embedding = OpenAIEmbeddings(deployment=azure_model, model=azure_model) - else: - embedding = OpenAIEmbeddings() - fs = LocalFileStore(config.local_skill_library_vectordb_path) - cached_embedding = CacheBackedEmbeddings.from_bytes_store( - embedding, fs, namespace=embedding.model - ) - return cached_embedding diff --git a/creator/memgpt/__init__.py b/creator/memgpt/__init__.py index 729aa6a..c87e56a 100644 --- a/creator/memgpt/__init__.py +++ b/creator/memgpt/__init__.py @@ -1,8 +1,6 @@ from .agent import create_memgpt -from .config import memory_config __all__ = [ - "create_memgpt", - "memory_config" + "create_memgpt" ] diff --git a/creator/memgpt/config.py b/creator/memgpt/config.py deleted file mode 100644 index ad5d71b..0000000 --- a/creator/memgpt/config.py +++ /dev/null @@ -1,39 +0,0 @@ -import os -from .constants import ( - DEFAULT_PERSONA, - DEFAULT_HUMAN, - CORE_MEMORY_HUMAN_CHAR_LIMIT, - CORE_MEMORY_PERSONA_CHAR_LIMIT -) - - -class AttrDict(dict): - def __getattr__(self, key): - try: - return self[key] - except KeyError: - raise AttributeError(key) - - def __setattr__(self, key, value): - self[key] = value - - def __delattr__(self, key): - if key in self: - del self[key] - else: - raise AttributeError(key) - - -memory_config = AttrDict({ - "memory_path": os.path.expanduser("~/.cache/open_creator/memory"), - "system_prompt_path": os.path.join(os.path.dirname(__file__), "prompts", "system_prompt.md"), - "function_schema_path": os.path.join(os.path.dirname(__file__), "prompts", "function_schema.json"), - # replace with your own session_id - "session_id": None, - "persona": DEFAULT_PERSONA, - "human": DEFAULT_HUMAN, - "persona_char_limit": CORE_MEMORY_PERSONA_CHAR_LIMIT, - "human_char_limit": CORE_MEMORY_HUMAN_CHAR_LIMIT, - "page_size": 5, - "use_vector_search": True, -}) diff --git a/creator/memgpt/constants.py b/creator/memgpt/constants.py index ffdd844..01cc533 100644 --- a/creator/memgpt/constants.py +++ b/creator/memgpt/constants.py @@ -1,29 +1,3 @@ -DEFAULT_PERSONA = """ -The following is a blank slate starter persona, I need to expand this to develop my own personality. - -My name is MemGPT. -I am kind, thoughtful, and inquisitive. -""" - - -DEFAULT_HUMAN = """ -This is what I know so far about the user, I should expand this as I learn more about them. - -First name: Chad -Last name: ? -Gender: Male -Age: ? -Nationality: ? -Occupation: Computer science PhD student at UC Berkeley -Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO -""" - -DEFAULT_AGENT_SUBTASKS = """ - - create/save/search skill - - run/test/refactor skill - - show skill -""" - # the number of tokens consumed in a call before a system warning goes to the agent MESSAGE_SUMMARY_WARNING_TOKENS = 6000 # Default memory limits @@ -34,5 +8,4 @@ FUNC_FAILED_HEARTBEAT_MESSAGE = "Function call failed" REQ_HEARTBEAT_MESSAGE = "request_heartbeat == true" -MESSAGE_CHATGPT_FUNCTION_MODEL = "gpt-3.5-turbo" MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE = "You are a helpful assistant. Keep your responses short and concise." diff --git a/creator/memgpt/memory/__init__.py b/creator/memgpt/memory/__init__.py index 734c7b9..2c1e571 100644 --- a/creator/memgpt/memory/__init__.py +++ b/creator/memgpt/memory/__init__.py @@ -1,7 +1,7 @@ -from .schema import build_memory from .core_memory import CoreMemory from .recall_memory import RecallMemory from .archival_memory import ArchivalMemory +from .builder import build_memory from ..time_utils import get_local_time diff --git a/creator/memgpt/memory/archival_memory.py b/creator/memgpt/memory/archival_memory.py index 0d3865b..e04a526 100644 --- a/creator/memgpt/memory/archival_memory.py +++ b/creator/memgpt/memory/archival_memory.py @@ -1,4 +1,4 @@ -from .schema import ArchivalMessage +from .converter import ArchivalMessage from .recall_memory import RecallMemory diff --git a/creator/memgpt/memory/builder.py b/creator/memgpt/memory/builder.py new file mode 100644 index 0000000..4dc32cd --- /dev/null +++ b/creator/memgpt/memory/builder.py @@ -0,0 +1,15 @@ +import uuid + +from langchain.memory.chat_message_histories import SQLChatMessageHistory + +from .converter import MessageConverter + + +def build_memory(memory_path:str, session_id=None): + if session_id is None: + session_id = str(uuid.uuid4()) + return SQLChatMessageHistory( + session_id=session_id, + connection_string=f"sqlite:///{memory_path}/.langchain.db", + custom_message_converter=MessageConverter() + ) diff --git a/creator/memgpt/memory/converter.py b/creator/memgpt/memory/converter.py new file mode 100644 index 0000000..35fb609 --- /dev/null +++ b/creator/memgpt/memory/converter.py @@ -0,0 +1,64 @@ +from typing import Any +import datetime + +from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, FunctionMessage +from langchain.memory.chat_message_histories.sql import BaseMessageConverter + +from ..time_utils import get_local_time +from .schema import MemoryMessage + + +class ArchivalMessage(BaseMessage): + type: str = "archival" + + +class MessageConverter(BaseMessageConverter): + + def from_sql_model(self, sql_message: Any) -> BaseMessage: + created_at = sql_message.created_at.strftime("%Y-%m-%d %I:%M:%S %p %Z%z") + sql_message.additional_kwargs.update({"created_at": created_at}) + if sql_message.type == 'human': + return HumanMessage( + content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs + ) + elif "ai" in sql_message.type.lower(): + return AIMessage( + content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs + ) + elif "system" in sql_message.type.lower(): + return SystemMessage( + content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs + ) + elif "function" in sql_message.type.lower(): + return FunctionMessage( + content=sql_message.content, + name=sql_message.additional_kwargs.get("name", ""), + additional_kwargs=sql_message.additional_kwargs + ) + elif "archival" in sql_message.type.lower(): + return ArchivalMessage( + content=sql_message.content, + additional_kwargs=sql_message.additional_kwargs + ) + else: + raise ValueError(f'Unknown message type: {sql_message.type}') + + def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: + now = get_local_time() + now_datetime = datetime.datetime.strptime(now, "%Y-%m-%d %I:%M:%S %p %Z%z") + if isinstance(message, FunctionMessage): + message.additional_kwargs.update({"name": message.name}) + message.additional_kwargs.update({"created_at": now}) + return MemoryMessage( + session_id=session_id, + type=message.type, + content=message.content, + created_at=now_datetime, + additional_kwargs=message.additional_kwargs + ) + + def get_sql_model_class(self) -> Any: + return MemoryMessage diff --git a/creator/memgpt/memory/recall_memory.py b/creator/memgpt/memory/recall_memory.py index 395e174..e9c18c1 100644 --- a/creator/memgpt/memory/recall_memory.py +++ b/creator/memgpt/memory/recall_memory.py @@ -1,9 +1,7 @@ -import numpy as np - -from creator.llm import create_embedding -from creator.config.library import config -from creator.retrivever.score_functions import cosine_similarity from langchain.memory.chat_message_histories import SQLChatMessageHistory +from langchain.adapters.openai import convert_message_to_dict + +from creator.retrivever.memory_retrivever import MemoryVectorStore from .base import BaseMemory @@ -18,12 +16,11 @@ class RecallMemory(BaseMemory): effectively allowing it to 'remember' prior engagements with a user. """ - def __init__(self, message_database: SQLChatMessageHistory, use_vector_search=False): + def __init__(self, message_database: SQLChatMessageHistory, use_vector_search: bool = True): self.message_database = message_database self.use_vector_search = use_vector_search - if use_vector_search: - self.embeddings = dict() - self.embedding_model = create_embedding(config) + if self.use_vector_search: + self.retrivever = MemoryVectorStore() def __len__(self): return len(self.message_database.messages) @@ -54,12 +51,6 @@ def _filter_messages(self): """Utility to filter messages based on roles.""" return [d for d in self.message_database.messages if d.type not in ['system', 'function', 'archival']] - async def _get_or_compute_embedding(self, message_str): - """Retrieve or compute the embedding for a given string.""" - if message_str not in self.embeddings: - self.embeddings[message_str] = self.embedding_model.embed_query(message_str) - return self.embeddings[message_str] - async def search(self, query, page, start_date=None, end_date=None): """Simple text-based search""" matches = self._filter_messages() @@ -70,11 +61,11 @@ async def search(self, query, page, start_date=None, end_date=None): if query: if self.use_vector_search: - message_pool_filtered = [d for d in matches if await self._get_or_compute_embedding(d.content)] - query_vec = self.embedding_model.embed_query(query) - docs_matrix = np.array([self.embeddings[d.content] for d in message_pool_filtered]) - indexes, scores = cosine_similarity(docs_matrix=docs_matrix, query_vec=query_vec, k=len(docs_matrix)) - matches = [message_pool_filtered[i] for i in indexes] + texts = [d.content for d in matches] + metadatas = [convert_message_to_dict(match) for match in matches] + self.retrivever.index(documents=texts, metadatas=metadatas) + matches = self.retrivever.search(query=query, top_k=len(matches)) + self.retrivever.reset() else: matches = [d for d in matches if d.content and query.lower() in d.content.lower()] return self._paginate_results(matches, page) diff --git a/creator/memgpt/memory/schema.py b/creator/memgpt/memory/schema.py index 7791615..45c6986 100644 --- a/creator/memgpt/memory/schema.py +++ b/creator/memgpt/memory/schema.py @@ -1,92 +1,34 @@ -from typing import Any -import datetime import uuid +import datetime from sqlalchemy.orm import declarative_base from sqlalchemy import Column, Integer, Text, DateTime, JSON -from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, FunctionMessage -from langchain.memory.chat_message_histories.sql import BaseMessageConverter -from langchain.memory.chat_message_histories import SQLChatMessageHistory - -from ..time_utils import get_local_time - Base = declarative_base() -class ArchivalMessage(BaseMessage): - type: str = "archival" - - class MemoryMessage(Base): - __tablename__ = 'memory_messages' + __tablename__ = "memory_messages" id = Column(Integer, primary_key=True) session_id = Column(Text) type = Column(Text) content = Column(Text) - created_at = Column(DateTime) + created_at = Column(DateTime, default=datetime.datetime.now) additional_kwargs = Column(JSON) -class MessageConverter(BaseMessageConverter): +class SessionConfig(Base): + __tablename__ = "session_configs" - def from_sql_model(self, sql_message: Any) -> BaseMessage: - created_at = sql_message.created_at.strftime("%Y-%m-%d %I:%M:%S %p %Z%z") - sql_message.additional_kwargs.update({"created_at": created_at}) - if sql_message.type == 'human': - return HumanMessage( - content=sql_message.content, - additional_kwargs=sql_message.additional_kwargs - ) - elif "ai" in sql_message.type.lower(): - return AIMessage( - content=sql_message.content, - additional_kwargs=sql_message.additional_kwargs - ) - elif "system" in sql_message.type.lower(): - return SystemMessage( - content=sql_message.content, - additional_kwargs=sql_message.additional_kwargs - ) - elif "function" in sql_message.type.lower(): - return FunctionMessage( - content=sql_message.content, - name=sql_message.additional_kwargs.get("name", ""), - additional_kwargs=sql_message.additional_kwargs - ) - elif "archival" in sql_message.type.lower(): - return ArchivalMessage( - content=sql_message.content, - additional_kwargs=sql_message.additional_kwargs - ) - else: - raise ValueError(f'Unknown message type: {sql_message.type}') - - def to_sql_model(self, message: BaseMessage, session_id: str) -> Any: - now = get_local_time() - now_datetime = datetime.datetime.strptime(now, "%Y-%m-%d %I:%M:%S %p %Z%z") - if isinstance(message, FunctionMessage): - message.additional_kwargs.update({"name": message.name}) - message.additional_kwargs.update({"created_at": now}) - return MemoryMessage( - session_id=session_id, - type=message.type, - content=message.content, - created_at=now_datetime, - additional_kwargs=message.additional_kwargs - ) - - def get_sql_model_class(self) -> Any: - return MemoryMessage - - -def build_memory(memory_path:str, session_id=None): - if session_id is None: - session_id = str(uuid.uuid4()) - return SQLChatMessageHistory( - session_id=session_id, - connection_string=f"sqlite:///{memory_path}/.langchain.db", - custom_message_converter=MessageConverter() - ) + id = Column(Integer, primary_key=True) + session_id = Column(Text, default=lambda: str(uuid.uuid4())) + title = Column(Text, default="Untitled") + created_at = Column(DateTime, default=datetime.datetime.now) + model_name = Column(Text, default="gpt-4") + persona = Column(Text) + human = Column(Text) + persona_char_limit = Column(Integer, default=2000) + human_char_limit = Column(Integer, default=2000) + page_size = Column(Integer, default=5) diff --git a/creator/memgpt/prompts/function_schema.json b/creator/prompts/memgpt_function_schema.json similarity index 100% rename from creator/memgpt/prompts/function_schema.json rename to creator/prompts/memgpt_function_schema.json diff --git a/creator/memgpt/prompts/system_prompt.md b/creator/prompts/memgpt_system_prompt.md similarity index 100% rename from creator/memgpt/prompts/system_prompt.md rename to creator/prompts/memgpt_system_prompt.md diff --git a/creator/retrivever/base.py b/creator/retrivever/base.py index ab04261..42cfc62 100644 --- a/creator/retrivever/base.py +++ b/creator/retrivever/base.py @@ -1,105 +1,44 @@ -import numpy as np -from typing import List -import json -import os - -from creator.llm import create_embedding -from creator.config.library import config - -from .score_functions import cosine_similarity +from typing import List, Any, Dict +from langchain.vectorstores.qdrant import Qdrant +from langchain.docstore.document import Document +from creator.utils import generate_uuid_like_string class BaseVectorStore: - def __init__(self, skill_library_path: str = ""): - - self.vectordb_path: str = config.local_skill_library_vectordb_path - self.skill_library_path = config.local_skill_library_path - self.vector_store = {} - self.embeddings = None - self.embedding_model = create_embedding(config) - self.sorted_keys = [] - self.query_cache = {} - - if skill_library_path and os.path.exists(skill_library_path): - self.skill_library_path = skill_library_path - - if os.path.isdir(self.skill_library_path): - self.query_cache_path = self.vectordb_path + "/query_cache.json" - self.vectordb_path = self.vectordb_path + "/vector_db.json" - if os.path.exists(self.query_cache_path): - with open(self.query_cache_path, mode="r", encoding="utf-8") as f: - self.query_cache = json.load(f) - - if os.path.exists(self.vectordb_path): - # load vectordb - with open(self.vectordb_path, mode="r", encoding="utf-8") as f: - self.vector_store = json.load(f) - - self.update_index() - - def update_index(self): - # glob skill_library_path to find `embedding_text.txt` - embeddings = [] - - for root, dirs, files in os.walk(self.skill_library_path): - for file in files: - if root not in self.vector_store and file == "embedding_text.txt": - embedding_text_path = os.path.join(root, file) - with open(embedding_text_path, mode="r", encoding="utf-8") as f: - embedding_text = f.read() - - skill_path = os.path.join(root, "skill.json") - with open(skill_path, encoding="utf-8") as f: - skill_json = json.load(f) - skill_json["skill_id"] = root - skill_json["embedding_text"] = embedding_text - self.vector_store[root] = skill_json - - # index embedding_texts - no_embedding_obj = {key:value for key, value in self.vector_store.items() if "embedding" not in value} - if len(no_embedding_obj) > 0: - no_embedding_texts = [] - sorted_keys = sorted(no_embedding_obj) - for key in sorted_keys: - no_embedding_texts.append(no_embedding_obj[key]["embedding_text"]) - - embeddings = self.embedding_model.embed_documents(no_embedding_texts) - for i, key in enumerate(sorted_keys): - self.vector_store[key]["embedding"] = embeddings[i] - - self.sorted_keys = sorted(self.vector_store) - embeddings = [] - for key in self.sorted_keys: - embeddings.append(self.vector_store[key]["embedding"]) - self.embeddings = np.array(embeddings) - # save to vectordb - with open(self.vectordb_path, "w", encoding="utf-8") as f: - json.dump(self.vector_store, f) - - def save_query_cache(self): - with open(self.query_cache_path, "w", encoding="utf-8") as f: - json.dump(self.query_cache, f) + def __init__(self, vectordb_path, embedding, collection_name): + self.vectordb_path = vectordb_path + self.embedding = embedding + self.collection_name = collection_name + self.db = None + + def _preprocess(self, doc: Any, **kwargs): + """Preprocess the input doc into text""" + return doc + + def _postprocess(self, documents: List[Document]): + """Postprocess the documents""" + return documents + + def _update_index(self): + pass + + def reset(self): + self.db = None + + def index(self, documents: List[Any], ids: List[str] = None, metadatas: List[Dict] = None): + """Public method to index a document.""" + if metadatas is None: + metadatas = documents + texts = [self._preprocess(doc) for doc in documents] + if ids is None: + ids = [generate_uuid_like_string() for _ in range(len(texts))] + if self.db is None: + self.db = Qdrant.from_texts(texts=texts, embedding=self.embedding, metadatas=metadatas, ids=ids, path=self.vectordb_path, collection_name=self.collection_name) + else: + self.db.add_texts(texts=texts, metadatas=metadatas, ids=ids) def search(self, query: str, top_k: int = 3, threshold=0.8) -> List[dict]: - key = str((query, top_k, threshold)) - if key in self.query_cache: - return self.query_cache[key] - - self.update_index() - - query_embedding = self.embedding_model.embed_query(query) - query_embedding = np.array(query_embedding) - indexes, scores = cosine_similarity(docs_matrix=self.embeddings, query_vec=query_embedding, k=top_k) - results = [] - for i, index in enumerate(indexes): - if scores[i] < threshold: - break - result = self.vector_store[self.sorted_keys[index]] - result = result.copy() - result.pop("embedding") - result["score"] = scores[i] - results.append(result) - self.query_cache[key] = results - self.save_query_cache() - return results + self._update_index() + documents = self.db.similarity_search(query=query, k=top_k, score_threshold=threshold) + return self._postprocess(documents) diff --git a/creator/retrivever/embedding_creator.py b/creator/retrivever/embedding_creator.py new file mode 100644 index 0000000..044ef45 --- /dev/null +++ b/creator/retrivever/embedding_creator.py @@ -0,0 +1,20 @@ +import os +from langchain.embeddings import OpenAIEmbeddings, CacheBackedEmbeddings +from langchain.storage import LocalFileStore + + +def create_embedding(config): + + use_azure = True if os.getenv("OPENAI_API_TYPE", None) == "azure" else False + + if use_azure: + azure_model = os.getenv("EMBEDDING_DEPLOYMENT_NAME", None) + print(azure_model) + embedding = OpenAIEmbeddings(deployment=azure_model, model=azure_model) + else: + embedding = OpenAIEmbeddings() + fs = LocalFileStore(config.embedding_cache_path) + cached_embedding = CacheBackedEmbeddings.from_bytes_store( + embedding, fs, namespace=embedding.model + ) + return cached_embedding diff --git a/creator/retrivever/memory_retrivever.py b/creator/retrivever/memory_retrivever.py new file mode 100644 index 0000000..b5aef32 --- /dev/null +++ b/creator/retrivever/memory_retrivever.py @@ -0,0 +1,21 @@ +from typing import List + +from langchain.docstore.document import Document +from langchain.adapters.openai import convert_openai_messages + +from creator.config.library import config + +from .base import BaseVectorStore +from .embedding_creator import create_embedding + + +class MemoryVectorStore(BaseVectorStore): + + def __init__(self, collection_name="recall_memory"): + self.vectordb_path = config.vectordb_path + self.embedding = create_embedding(config) + self.collection_name = collection_name + self.db = None + + def _postprocess(self, documents: List[Document]): + return [convert_openai_messages(doc.metadata) for doc in documents] diff --git a/creator/retrivever/score_functions.py b/creator/retrivever/score_functions.py deleted file mode 100644 index 2d0f5d6..0000000 --- a/creator/retrivever/score_functions.py +++ /dev/null @@ -1,7 +0,0 @@ -import numpy as np - - -def cosine_similarity(docs_matrix, query_vec, k=3): - similarities = np.dot(docs_matrix, query_vec) / (np.linalg.norm(docs_matrix, axis=1) * np.linalg.norm(query_vec)) - top_k_indices = np.argsort(similarities)[-k:][::-1] - return top_k_indices, similarities[top_k_indices] diff --git a/creator/retrivever/skill_retrivever.py b/creator/retrivever/skill_retrivever.py new file mode 100644 index 0000000..97c93db --- /dev/null +++ b/creator/retrivever/skill_retrivever.py @@ -0,0 +1,39 @@ +from typing import List +import json +import os + +from langchain.docstore.document import Document + +from creator.config.library import config + +from .base import BaseVectorStore +from .embedding_creator import create_embedding + + +class SkillVectorStore(BaseVectorStore): + + def __init__(self): + self.vectordb_path = config.vectordb_path + self.embedding = create_embedding(config) + self.collection_name = "skill_library" + self.db = None + + def _update_index(self): + # glob skill_library_path to find `embedding_text.txt` + texts = [] + metadatas = [] + for root, dirs, files in os.walk(config.local_skill_library_path): + for file in files: + if file == "embedding_text.txt": + embedding_text_path = os.path.join(root, file) + with open(embedding_text_path, mode="r", encoding="utf-8") as f: + embedding_text = f.read() + skill_path = os.path.join(root, "skill.json") + with open(skill_path, encoding="utf-8") as f: + skill_json = json.load(f) + texts.append(embedding_text) + metadatas.append(skill_json) + self.index(documents=texts, metadatas=metadatas) + + def _postprocess(self, documents: List[Document]): + return [doc.metadata for doc in documents] diff --git a/creator/skill_library/open-creator/create/conversation_history.json b/creator/skill_library/open-creator/create/conversation_history.json deleted file mode 100644 index fbf893c..0000000 --- a/creator/skill_library/open-creator/create/conversation_history.json +++ /dev/null @@ -1,6 +0,0 @@ -[ - { - "role": "user", - "content": "# file name: create.py\nimport creator\nfrom creator.schema.skill import CodeSkill\nfrom typing import Optional, List\n\n\ndef create(\n request: Optional[str] = None,\n messages: Optional[List[dict]] = None,\n messages_json_path: Optional[str] = None,\n skill_path: Optional[str] = None,\n skill_json_path: Optional[str] = None,\n file_content: Optional[str] = None,\n file_path: Optional[str] = None,\n huggingface_repo_id: Optional[str] = None,\n huggingface_skill_path: Optional[str] = None,\n) -> CodeSkill:\n \"\"\"Create a skill from various sources.\n\n Args:\n request (Optional[str], optional): Request string. Defaults to None.\n messages (Optional[List[dict]], optional): Messages in list of dict format. Defaults to None.\n messages_json_path (Optional[str], optional): Path to messages JSON file. Defaults to None.\n skill_path (Optional[str], optional): Path to skill directory. Defaults to None.\n skill_json_path (Optional[str], optional): Path to skill JSON file. Defaults to None.\n file_content (Optional[str], optional): File content. Defaults to None.\n file_path (Optional[str], optional): Path to file. Defaults to None.\n huggingface_repo_id (Optional[str], optional): Huggingface repo ID. Defaults to None.\n huggingface_skill_path (Optional[str], optional): Huggingface skill path. Defaults to None.\n\n Returns:\n CodeSkill: Created skill\n Example:\n >>> skill = creator.create(request=\"filter how many prime numbers are in 201\")\n >>> skill = creator.create(messages=[{\"role\": \"user\",\"content\": \"write a program to list all the python functions and their docstrings in a directory\"},{\"role\": \"assistant\",\"content\": \"Sure, I can help with that. Here's the plan:\\n\\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\\n4. Finally, we will print out the function names and their docstrings.\\n\\nLet's start with step 1: getting a list of all Python files in the specified directory.\",\"function_call\": {\"name\": \"run_code\",\"arguments\": \"{\\n \\\"language\\\": \\\"python\\\",\\n \\\"code\\\": \\\"import os\\\\nimport glob\\\\n\\\\n# Get the current working directory\\\\ncwd = os.getcwd()\\\\n\\\\n# Get a list of all Python files in the directory\\\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\\\n\\\\npython_files\\\"\\n}\"}}])\n >>> skill = creator.create(messages_json_path=\"./messages_example.json\")\n >>> skill = creator.create(file_path=\"../creator/utils/ask_human.py\")\n >>> skill = creator.create(huggingface_repo_id=\"Sayoyo/skill-library\", huggingface_skill_path=\"extract_pdf_section\")\n >>> skill = creator.create(skill_json_path=os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/create/skill.json\")\n \"\"\"\n if request is not None:\n skill = creator.create(request=request)\n elif messages is not None:\n skill = creator.create(messages=messages)\n elif messages_json_path is not None:\n skill = creator.create(messages_json_path=messages_json_path)\n elif skill_path is not None:\n skill = creator.create(skill_path=skill_path)\n elif skill_json_path is not None:\n skill = creator.create(skill_json_path=skill_json_path)\n elif file_content is not None:\n skill = creator.create(file_content=file_content)\n elif file_path is not None:\n skill = creator.create(file_path=file_path)\n elif huggingface_repo_id is not None and huggingface_skill_path is not None:\n skill = creator.create(\n huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path\n )\n else:\n raise ValueError(\"At least one argument must be provided.\")\n\n return skill\n" - } -] \ No newline at end of file diff --git a/creator/skill_library/open-creator/create/embedding_text.txt b/creator/skill_library/open-creator/create/embedding_text.txt deleted file mode 100644 index 130fb4d..0000000 --- a/creator/skill_library/open-creator/create/embedding_text.txt +++ /dev/null @@ -1,22 +0,0 @@ -create -Create a skill from various sources. - Args: - request (Optional[str], optional): Request string. Defaults to None. - messages (Optional[List[dict]], optional): Messages in list of dict format. Defaults to None. - messages_json_path (Optional[str], optional): Path to messages JSON file. Defaults to None. - skill_path (Optional[str], optional): Path to skill directory. Defaults to None. - skill_json_path (Optional[str], optional): Path to skill JSON file. Defaults to None. - file_content (Optional[str], optional): File content. Defaults to None. - file_path (Optional[str], optional): Path to file. Defaults to None. - huggingface_repo_id (Optional[str], optional): Huggingface repo ID. Defaults to None. - huggingface_skill_path (Optional[str], optional): Huggingface skill path. Defaults to None. - - Returns: - CodeSkill: Created skill - Example: - >>> skill = create(request="filter how many prime numbers are in 201") - >>> skill = create(messages=[{"role": "user","content": "write a program to list all the python functions and their docstrings in a directory"},{"role": "assistant","content": "Sure, I can help with that. Here's the plan:\n\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\n4. Finally, we will print out the function names and their docstrings.\n\nLet's start with step 1: getting a list of all Python files in the specified directory.","function_call": {"name": "run_code","arguments": "{\n \"language\": \"python\",\n \"code\": \"import os\\nimport glob\\n\\n# Get the current working directory\\ncwd = os.getcwd()\\n\\n# Get a list of all Python files in the directory\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\n\\npython_files\"\n}"}}]) - >>> skill = create(messages_json_path="./messages_example.json") - >>> skill = create(file_path="../creator/utils/ask_human.py") - >>> skill = create(huggingface_repo_id="Sayoyo/skill-library", huggingface_skill_path="extract_pdf_section") - >>> skill = create(skill_json_path=os.path.expanduser("~") + "/.cache/open_creator/skill_library/create/skill.json") \ No newline at end of file diff --git a/creator/skill_library/open-creator/create/function_call.json b/creator/skill_library/open-creator/create/function_call.json deleted file mode 100644 index 98ec578..0000000 --- a/creator/skill_library/open-creator/create/function_call.json +++ /dev/null @@ -1,46 +0,0 @@ -{ - "name": "create", - "description": "Create a skill from various sources.\n\nskill = creator.create(request=\"filter how many prime numbers are in 201\")", - "parameters": { - "type": "object", - "properties": { - "request": { - "type": "string", - "description": "Request string." - }, - "messages": { - "type": "array", - "description": "Messages in list of dict format." - }, - "messages_json_path": { - "type": "string", - "description": "Path to messages JSON file." - }, - "skill_path": { - "type": "string", - "description": "Path to skill directory." - }, - "skill_json_path": { - "type": "string", - "description": "Path to skill JSON file." - }, - "file_content": { - "type": "string", - "description": "File content." - }, - "file_path": { - "type": "string", - "description": "Path to file." - }, - "huggingface_repo_id": { - "type": "string", - "description": "Huggingface repo ID." - }, - "huggingface_skill_path": { - "type": "string", - "description": "Huggingface skill path." - } - }, - "required": [] - } -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/create/install_dependencies.sh b/creator/skill_library/open-creator/create/install_dependencies.sh deleted file mode 100644 index c16ae4c..0000000 --- a/creator/skill_library/open-creator/create/install_dependencies.sh +++ /dev/null @@ -1 +0,0 @@ -pip install -U "open-creator" diff --git a/creator/skill_library/open-creator/create/skill.json b/creator/skill_library/open-creator/create/skill.json deleted file mode 100644 index 64414cb..0000000 --- a/creator/skill_library/open-creator/create/skill.json +++ /dev/null @@ -1,106 +0,0 @@ -{ - "skill_name": "create", - "skill_description": "Create a skill from various sources.", - "skill_metadata": { - "created_at": "2023-10-03 22:39:34", - "author": "gongjunmin", - "updated_at": "2023-10-03 22:39:34", - "usage_count": 0, - "version": "1.0.0", - "additional_kwargs": {} - }, - "skill_tags": [ - "create", - "skill", - "source" - ], - "skill_usage_example": "skill = creator.create(request=\"filter how many prime numbers are in 201\")", - "skill_program_language": "python", - "skill_code": "from creator.core import creator\nfrom creator.core.skill import CodeSkill\nfrom typing import Optional, List\n\n\ndef create(\n request: Optional[str] = None,\n messages: Optional[List[dict]] = None,\n messages_json_path: Optional[str] = None,\n skill_path: Optional[str] = None,\n skill_json_path: Optional[str] = None,\n file_content: Optional[str] = None,\n file_path: Optional[str] = None,\n huggingface_repo_id: Optional[str] = None,\n huggingface_skill_path: Optional[str] = None,\n) -> CodeSkill:\n \"\"\"Create a skill from various sources.\n\n Args:\n request (Optional[str], optional): Request string. Defaults to None.\n messages (Optional[List[dict]], optional): Messages in list of dict format. Defaults to None.\n messages_json_path (Optional[str], optional): Path to messages JSON file. Defaults to None.\n skill_path (Optional[str], optional): Path to skill directory. Defaults to None.\n skill_json_path (Optional[str], optional): Path to skill JSON file. Defaults to None.\n file_content (Optional[str], optional): File content. Defaults to None.\n file_path (Optional[str], optional): Path to file. Defaults to None.\n huggingface_repo_id (Optional[str], optional): Huggingface repo ID. Defaults to None.\n huggingface_skill_path (Optional[str], optional): Huggingface skill path. Defaults to None.\n\n Returns:\n CodeSkill: Created skill\n Example:\n >>> skill = create(request=\"filter how many prime numbers are in 201\")\n >>> skill = create(messages=[{\"role\": \"user\",\"content\": \"write a program to list all the python functions and their docstrings in a directory\"},{\"role\": \"assistant\",\"content\": \"Sure, I can help with that. Here's the plan:\\n\\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\\n4. Finally, we will print out the function names and their docstrings.\\n\\nLet's start with step 1: getting a list of all Python files in the specified directory.\",\"function_call\": {\"name\": \"run_code\",\"arguments\": \"{\\n \\\"language\\\": \\\"python\\\",\\n \\\"code\\\": \\\"import os\\\\nimport glob\\\\n\\\\n# Get the current working directory\\\\ncwd = os.getcwd()\\\\n\\\\n# Get a list of all Python files in the directory\\\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\\\n\\\\npython_files\\\"\\n}\"}}])\n >>> skill = create(messages_json_path=\"./messages_example.json\")\n >>> skill = create(file_path=\"../creator/utils/ask_human.py\")\n >>> skill = create(huggingface_repo_id=\"Sayoyo/skill-library\", huggingface_skill_path=\"extract_pdf_section\")\n >>> skill = create(skill_json_path=os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/create/skill.json\")\n \"\"\"\n if request is not None:\n skill = creator.create(request=request)\n elif messages is not None:\n skill = creator.create(messages=messages)\n elif messages_json_path is not None:\n skill = creator.create(messages_json_path=messages_json_path)\n elif skill_path is not None:\n skill = creator.create(skill_path=skill_path)\n elif skill_json_path is not None:\n skill = creator.create(skill_json_path=skill_json_path)\n elif file_content is not None:\n skill = creator.create(file_content=file_content)\n elif file_path is not None:\n skill = creator.create(file_path=file_path)\n elif huggingface_repo_id is not None and huggingface_skill_path is not None:\n skill = creator.create(\n huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path\n )\n else:\n raise ValueError(\"At least one argument must be provided.\")\n\n return skill\n", - "skill_parameters": [ - { - "param_name": "request", - "param_type": "string", - "param_description": "Request string.", - "param_required": false, - "param_default": null - }, - { - "param_name": "messages", - "param_type": "array", - "param_description": "Messages in list of dict format.", - "param_required": false, - "param_default": null - }, - { - "param_name": "messages_json_path", - "param_type": "string", - "param_description": "Path to messages JSON file.", - "param_required": false, - "param_default": null - }, - { - "param_name": "skill_path", - "param_type": "string", - "param_description": "Path to skill directory.", - "param_required": false, - "param_default": null - }, - { - "param_name": "skill_json_path", - "param_type": "string", - "param_description": "Path to skill JSON file.", - "param_required": false, - "param_default": null - }, - { - "param_name": "file_content", - "param_type": "string", - "param_description": "File content.", - "param_required": false, - "param_default": null - }, - { - "param_name": "file_path", - "param_type": "string", - "param_description": "Path to file.", - "param_required": false, - "param_default": null - }, - { - "param_name": "huggingface_repo_id", - "param_type": "string", - "param_description": "Huggingface repo ID.", - "param_required": false, - "param_default": null - }, - { - "param_name": "huggingface_skill_path", - "param_type": "string", - "param_description": "Huggingface skill path.", - "param_required": false, - "param_default": null - } - ], - "skill_return": { - "param_name": "CodeSkill", - "param_type": "object", - "param_description": "Created skill", - "param_required": true, - "param_default": null - }, - "skill_dependencies": [ - { - "dependency_name": "open-creator", - "dependency_version": "latest", - "dependency_type": "pacakge" - } - ], - "conversation_history": [ - { - "role": "user", - "content": "# file name: create.py\nimport creator\nfrom creator.schema.skill import CodeSkill\nfrom typing import Optional, List\n\n\ndef create(\n request: Optional[str] = None,\n messages: Optional[List[dict]] = None,\n messages_json_path: Optional[str] = None,\n skill_path: Optional[str] = None,\n skill_json_path: Optional[str] = None,\n file_content: Optional[str] = None,\n file_path: Optional[str] = None,\n huggingface_repo_id: Optional[str] = None,\n huggingface_skill_path: Optional[str] = None,\n) -> CodeSkill:\n \"\"\"Create a skill from various sources.\n\n Args:\n request (Optional[str], optional): Request string. Defaults to None.\n messages (Optional[List[dict]], optional): Messages in list of dict format. Defaults to None.\n messages_json_path (Optional[str], optional): Path to messages JSON file. Defaults to None.\n skill_path (Optional[str], optional): Path to skill directory. Defaults to None.\n skill_json_path (Optional[str], optional): Path to skill JSON file. Defaults to None.\n file_content (Optional[str], optional): File content. Defaults to None.\n file_path (Optional[str], optional): Path to file. Defaults to None.\n huggingface_repo_id (Optional[str], optional): Huggingface repo ID. Defaults to None.\n huggingface_skill_path (Optional[str], optional): Huggingface skill path. Defaults to None.\n\n Returns:\n CodeSkill: Created skill\n Example:\n >>> skill = creator.create(request=\"filter how many prime numbers are in 201\")\n >>> skill = creator.create(messages=[{\"role\": \"user\",\"content\": \"write a program to list all the python functions and their docstrings in a directory\"},{\"role\": \"assistant\",\"content\": \"Sure, I can help with that. Here's the plan:\\n\\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\\n4. Finally, we will print out the function names and their docstrings.\\n\\nLet's start with step 1: getting a list of all Python files in the specified directory.\",\"function_call\": {\"name\": \"run_code\",\"arguments\": \"{\\n \\\"language\\\": \\\"python\\\",\\n \\\"code\\\": \\\"import os\\\\nimport glob\\\\n\\\\n# Get the current working directory\\\\ncwd = os.getcwd()\\\\n\\\\n# Get a list of all Python files in the directory\\\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\\\n\\\\npython_files\\\"\\n}\"}}])\n >>> skill = creator.create(messages_json_path=\"./messages_example.json\")\n >>> skill = creator.create(file_path=\"../creator/utils/ask_human.py\")\n >>> skill = creator.create(huggingface_repo_id=\"Sayoyo/skill-library\", huggingface_skill_path=\"extract_pdf_section\")\n >>> skill = creator.create(skill_json_path=os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/create/skill.json\")\n \"\"\"\n if request is not None:\n skill = creator.create(request=request)\n elif messages is not None:\n skill = creator.create(messages=messages)\n elif messages_json_path is not None:\n skill = creator.create(messages_json_path=messages_json_path)\n elif skill_path is not None:\n skill = creator.create(skill_path=skill_path)\n elif skill_json_path is not None:\n skill = creator.create(skill_json_path=skill_json_path)\n elif file_content is not None:\n skill = creator.create(file_content=file_content)\n elif file_path is not None:\n skill = creator.create(file_path=file_path)\n elif huggingface_repo_id is not None and huggingface_skill_path is not None:\n skill = creator.create(\n huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path\n )\n else:\n raise ValueError(\"At least one argument must be provided.\")\n\n return skill\n" - } - ], - "test_summary": null -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/create/skill_code.py b/creator/skill_library/open-creator/create/skill_code.py deleted file mode 100644 index c722a4c..0000000 --- a/creator/skill_library/open-creator/create/skill_code.py +++ /dev/null @@ -1,61 +0,0 @@ -from creator.core import creator -from creator.core.skill import CodeSkill -from typing import Optional, List - - -def create( - request: Optional[str] = None, - messages: Optional[List[dict]] = None, - messages_json_path: Optional[str] = None, - skill_path: Optional[str] = None, - skill_json_path: Optional[str] = None, - file_content: Optional[str] = None, - file_path: Optional[str] = None, - huggingface_repo_id: Optional[str] = None, - huggingface_skill_path: Optional[str] = None, -) -> CodeSkill: - """Create a skill from various sources. - - Args: - request (Optional[str], optional): Request string. Defaults to None. - messages (Optional[List[dict]], optional): Messages in list of dict format. Defaults to None. - messages_json_path (Optional[str], optional): Path to messages JSON file. Defaults to None. - skill_path (Optional[str], optional): Path to skill directory. Defaults to None. - skill_json_path (Optional[str], optional): Path to skill JSON file. Defaults to None. - file_content (Optional[str], optional): File content. Defaults to None. - file_path (Optional[str], optional): Path to file. Defaults to None. - huggingface_repo_id (Optional[str], optional): Huggingface repo ID. Defaults to None. - huggingface_skill_path (Optional[str], optional): Huggingface skill path. Defaults to None. - - Returns: - CodeSkill: Created skill - Example: - >>> skill = create(request="filter how many prime numbers are in 201") - >>> skill = create(messages=[{"role": "user","content": "write a program to list all the python functions and their docstrings in a directory"},{"role": "assistant","content": "Sure, I can help with that. Here's the plan:\n\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\n4. Finally, we will print out the function names and their docstrings.\n\nLet's start with step 1: getting a list of all Python files in the specified directory.","function_call": {"name": "run_code","arguments": "{\n \"language\": \"python\",\n \"code\": \"import os\\nimport glob\\n\\n# Get the current working directory\\ncwd = os.getcwd()\\n\\n# Get a list of all Python files in the directory\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\n\\npython_files\"\n}"}}]) - >>> skill = create(messages_json_path="./messages_example.json") - >>> skill = create(file_path="../creator/utils/ask_human.py") - >>> skill = create(huggingface_repo_id="Sayoyo/skill-library", huggingface_skill_path="extract_pdf_section") - >>> skill = create(skill_json_path=os.path.expanduser("~") + "/.cache/open_creator/skill_library/create/skill.json") - """ - if request is not None: - skill = creator.create(request=request) - elif messages is not None: - skill = creator.create(messages=messages) - elif messages_json_path is not None: - skill = creator.create(messages_json_path=messages_json_path) - elif skill_path is not None: - skill = creator.create(skill_path=skill_path) - elif skill_json_path is not None: - skill = creator.create(skill_json_path=skill_json_path) - elif file_content is not None: - skill = creator.create(file_content=file_content) - elif file_path is not None: - skill = creator.create(file_path=file_path) - elif huggingface_repo_id is not None and huggingface_skill_path is not None: - skill = creator.create( - huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path - ) - else: - raise ValueError("At least one argument must be provided.") - - return skill diff --git a/creator/skill_library/open-creator/create/skill_doc.md b/creator/skill_library/open-creator/create/skill_doc.md deleted file mode 100644 index d134b0b..0000000 --- a/creator/skill_library/open-creator/create/skill_doc.md +++ /dev/null @@ -1,26 +0,0 @@ -## Skill Details: -- **Name**: create -- **Description**: Create a skill from various sources. -- **Version**: 1.0.0 -- **Usage**: -```python -skill = create(request="filter how many prime numbers are in 201") -skill = create(messages=[{"role": "user","content": "write a program to list all the python functions and their docstrings in a directory"},{"role": "assistant","content": "Sure, I can help with that. Here's the plan:\n\n1. First, we need to get a list of all Python files in the specified directory. We can do this by using the `os` and `glob` modules in Python.\n2. Then, for each Python file, we will parse the file to find all function definitions. We can do this by using the `ast` module in Python, which can parse Python source code into an abstract syntax tree (AST).\n3. For each function definition, we will extract the function's name and its docstring. The `ast` module can also help us with this.\n4. Finally, we will print out the function names and their docstrings.\n\nLet's start with step 1: getting a list of all Python files in the specified directory.","function_call": {"name": "run_code","arguments": "{\n \"language\": \"python\",\n \"code\": \"import os\\nimport glob\\n\\n# Get the current working directory\\ncwd = os.getcwd()\\n\\n# Get a list of all Python files in the directory\\npython_files = glob.glob(os.path.join(cwd, '*.py'))\\n\\npython_files\"\n}"}}]) -skill = create(messages_json_path="./messages_example.json") -skill = create(file_path="../creator/utils/ask_human.py") -skill = create(huggingface_repo_id="Sayoyo/skill-library", huggingface_skill_path="extract_pdf_section") -skill = create(skill_json_path=os.path.expanduser("~") + "/.cache/open_creator/skill_library/create/skill.json") -``` -- **Parameters**: - - **request** (string): Request string. - - **messages** (array): Messages in list of dict format. - - **messages_json_path** (string): Path to messages JSON file. - - **skill_path** (string): Path to skill directory. - - **skill_json_path** (string): Path to skill JSON file. - - **file_content** (string): File content. - - **file_path** (string): Path to file. - - **huggingface_repo_id** (string): Huggingface repo ID. - - **huggingface_skill_path** (string): Huggingface skill path. - -- **Returns**: - - **CodeSkill** (object): Created skill \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/conversation_history.json b/creator/skill_library/open-creator/save/conversation_history.json deleted file mode 100644 index 0ecf2e1..0000000 --- a/creator/skill_library/open-creator/save/conversation_history.json +++ /dev/null @@ -1,6 +0,0 @@ -[ - { - "role": "user", - "content": "# file name: save.py\nimport creator\nfrom creator.schema.skill import CodeSkill\n\n\ndef save(skill: CodeSkill, huggingface_repo_id: str = None, skill_path: str = None):\n \"\"\"\n Save a skill to a local path or a huggingface repo.\n \n Parameters:\n skill: CodeSkill object, the skill to be saved.\n huggingface_repo_id: str, optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo.\n skill_path: str, optional, the local path. If provided, the skill will be saved to this path.\n \n Returns:\n None\n \n Usage examples:\n ```python\n >>> import creator\n >>> import os\n >>> skill_json_path = os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json\"\n >>> skill = creator.create(skill_json_path=skill_json_path)\n >>> creator.save(skill=skill, huggingface_repo_id=\"ChuxiJ/skill_library\")\n ```\n or\n ```python\n >>> import creator\n >>> import os\n >>> skill_json_path = os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json\"\n >>> skill = creator.create(skill_json_path=skill_json_path)\n >>> creator.save(skill=skill, skill_path=\"/path/to/save\")\n ```\n \"\"\"\n if huggingface_repo_id is not None:\n creator.save_to_hub(skill=skill, huggingface_repo_id=huggingface_repo_id)\n elif skill_path is not None:\n creator.save_to_skill_path(skill=skill, skill_path=skill_path)\n else:\n raise ValueError(\"Either huggingface_repo_id or skill_path must be provided.\")\n \n" - } -] \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/embedding_text.txt b/creator/skill_library/open-creator/save/embedding_text.txt deleted file mode 100644 index 5ac09a5..0000000 --- a/creator/skill_library/open-creator/save/embedding_text.txt +++ /dev/null @@ -1,5 +0,0 @@ -save -Save a skill to a local path or a huggingface repo. -Usage examples: -save(skill=skill) or save(skill=skill, huggingface_repo_id='xxxx/skill_library') or save(skill=skill, skill_path='/path/to/save') -['save', 'skill', 'huggingface', 'local path'] \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/function_call.json b/creator/skill_library/open-creator/save/function_call.json deleted file mode 100644 index 7bf0f1d..0000000 --- a/creator/skill_library/open-creator/save/function_call.json +++ /dev/null @@ -1,24 +0,0 @@ -{ - "name": "save", - "description": "Save a skill to a local path or a huggingface repo.\n\nsave(skill=skill, huggingface_repo_id='xxxx/skill_library') or save(skill=skill, skill_path='/path/to/save')", - "parameters": { - "type": "object", - "properties": { - "skill": { - "type": "object", - "description": "CodeSkill object, the skill to be saved." - }, - "huggingface_repo_id": { - "type": "string", - "description": "optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo." - }, - "skill_path": { - "type": "string", - "description": "optional, the local path. If provided, the skill will be saved to this path." - } - }, - "required": [ - "skill" - ] - } -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/install_dependencies.sh b/creator/skill_library/open-creator/save/install_dependencies.sh deleted file mode 100644 index eb1ad2d..0000000 --- a/creator/skill_library/open-creator/save/install_dependencies.sh +++ /dev/null @@ -1 +0,0 @@ -pip install -U "open-creator" \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/skill.json b/creator/skill_library/open-creator/save/skill.json deleted file mode 100644 index af93afe..0000000 --- a/creator/skill_library/open-creator/save/skill.json +++ /dev/null @@ -1,59 +0,0 @@ -{ - "skill_name": "save", - "skill_description": "Save a skill to a local path or a huggingface repo.", - "skill_metadata": { - "created_at": "2023-10-04 09:54:43", - "author": "gongjunmin", - "updated_at": "2023-10-04 09:54:43", - "usage_count": 0, - "version": "1.0.0", - "additional_kwargs": {} - }, - "skill_tags": [ - "save", - "skill", - "huggingface", - "local path" - ], - "skill_usage_example": "save(skill=skill, huggingface_repo_id='ChuxiJ/skill_library') or save(skill=skill, skill_path='/path/to/save')", - "skill_program_language": "python", - "skill_code": "from creator.core import creator\nfrom creator.core.skill import CodeSkill\n\n\ndef save(skill: CodeSkill, huggingface_repo_id: str = None, skill_path: str = None):\n \"\"\"\n Save a skill to a local path or a huggingface repo.\n \n Parameters:\n skill: CodeSkill object, the skill to be saved.\n huggingface_repo_id: str, optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo.\n skill_path: str, optional, the local path. If provided, the skill will be saved to this path.\n \n Returns:\n None\n \n Example:\n >>> import creator\n >>> import os\n >>> skill_json_path = os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json\"\n >>> skill = creator.create(skill_json_path=skill_json_path)\n >>> save(skill=skill, huggingface_repo_id=\"ChuxiJ/skill_library\") # save to remote\n >>> save(skill=skill, skill_path=\"/path/to/save\") # save to local\n \"\"\"\n if huggingface_repo_id is not None:\n creator.save(skill=skill, huggingface_repo_id=huggingface_repo_id)\n elif skill_path is not None:\n creator.save(skill=skill, skill_path=skill_path)\n else:\n creator.save(skill=skill)", - "skill_parameters": [ - { - "param_name": "skill", - "param_type": "object", - "param_description": "CodeSkill object, the skill to be saved.", - "param_required": true, - "param_default": null - }, - { - "param_name": "huggingface_repo_id", - "param_type": "string", - "param_description": "optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo.", - "param_required": false, - "param_default": null - }, - { - "param_name": "skill_path", - "param_type": "string", - "param_description": "optional, the local path. If provided, the skill will be saved to this path.", - "param_required": false, - "param_default": null - } - ], - "skill_return": null, - "skill_dependencies": [ - { - "dependency_name": "open-creator", - "dependency_version": "latest", - "dependency_type": "package" - } - ], - "conversation_history": [ - { - "role": "user", - "content": "# file name: save.py\nimport creator\nfrom creator.schema.skill import CodeSkill\n\n\ndef save(skill: CodeSkill, huggingface_repo_id: str = None, skill_path: str = None):\n \"\"\"\n Save a skill to a local path or a huggingface repo.\n \n Parameters:\n skill: CodeSkill object, the skill to be saved.\n huggingface_repo_id: str, optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo.\n skill_path: str, optional, the local path. If provided, the skill will be saved to this path.\n \n Returns:\n None\n \n Usage examples:\n ```python\n >>> import creator\n >>> import os\n >>> skill_json_path = os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json\"\n >>> skill = creator.create(skill_json_path=skill_json_path)\n >>> creator.save(skill=skill, huggingface_repo_id=\"ChuxiJ/skill_library\")\n ```\n or\n ```python\n >>> import creator\n >>> import os\n >>> skill_json_path = os.path.expanduser(\"~\") + \"/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json\"\n >>> skill = creator.create(skill_json_path=skill_json_path)\n >>> creator.save(skill=skill, skill_path=\"/path/to/save\")\n ```\n \"\"\"\n if huggingface_repo_id is not None:\n creator.save_to_hub(skill=skill, huggingface_repo_id=huggingface_repo_id)\n elif skill_path is not None:\n creator.save_to_skill_path(skill=skill, skill_path=skill_path)\n else:\n raise ValueError(\"Either huggingface_repo_id or skill_path must be provided.\")\n \n" - } - ], - "test_summary": null -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/skill_code.py b/creator/skill_library/open-creator/save/skill_code.py deleted file mode 100644 index 01dfbdb..0000000 --- a/creator/skill_library/open-creator/save/skill_code.py +++ /dev/null @@ -1,30 +0,0 @@ -from creator.core import creator -from creator.core.skill import CodeSkill - - -def save(skill: CodeSkill, huggingface_repo_id: str = None, skill_path: str = None): - """ - Save a skill to a local path or a huggingface repo. - - Parameters: - skill: CodeSkill object, the skill to be saved. - huggingface_repo_id: str, optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo. - skill_path: str, optional, the local path. If provided, the skill will be saved to this path. - - Returns: - None - - Example: - >>> import creator - >>> import os - >>> skill_json_path = os.path.expanduser("~") + "/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json" - >>> skill = creator.create(skill_json_path=skill_json_path) - >>> save(skill=skill, huggingface_repo_id="ChuxiJ/skill_library") # save to remote - >>> save(skill=skill, skill_path="/path/to/save") # save to local - """ - if huggingface_repo_id is not None: - creator.save(skill=skill, huggingface_repo_id=huggingface_repo_id) - elif skill_path is not None: - creator.save(skill=skill, skill_path=skill_path) - else: - creator.save(skill=skill) \ No newline at end of file diff --git a/creator/skill_library/open-creator/save/skill_doc.md b/creator/skill_library/open-creator/save/skill_doc.md deleted file mode 100644 index 08b1c63..0000000 --- a/creator/skill_library/open-creator/save/skill_doc.md +++ /dev/null @@ -1,26 +0,0 @@ -## Skill Details: -- **Name**: save -- **Description**: Save a skill to a local path or a huggingface repo. -- **Version**: 1.0.0 -- **Usage**: -You need to create a skill first -```python -import creator -import os -skill_json_path = os.path.expanduser("~") + "/.cache/open_creator/skill_library/ask_run_code_confirm/skill.json" -skill = creator.create(skill_json_path=skill_json_path) -``` -```python -save(skill=skill, huggingface_repo_id="ChuxiJ/skill_library") -``` -or -```python -save(skill=skill, skill_path="/path/to/save") -``` -- **Parameters**: - - **skill** (object): CodeSkill object, the skill to be saved. - - Required: True - - **huggingface_repo_id** (string): optional, the ID of the huggingface repo. If provided, the skill will be saved to this repo. - - **skill_path** (string): optional, the local path. If provided, the skill will be saved to this path. - -- **Returns**: \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/conversation_history.json b/creator/skill_library/open-creator/search/conversation_history.json deleted file mode 100644 index 847ef4f..0000000 --- a/creator/skill_library/open-creator/search/conversation_history.json +++ /dev/null @@ -1,6 +0,0 @@ -[ - { - "role": "user", - "content": "# file name: search.py\nimport creator\nfrom creator.schema.skill import CodeSkill\n\n\ndef search(query: str, top_k=1, threshold=0.8) -> list[CodeSkill]:\n \"\"\"\n Search skills by query.\n \n Parameters:\n query: str, the query.\n top_k: int, optional, the maximum number of skills to return.\n threshold: float, optional, the minimum similarity score to return a skill.\n Returns:\n a list of CodeSkill objects.\n\n Example:\n >>> import creator\n >>> skills = search(\"I want to extract some pages from a pdf\")\n \"\"\"\n\n return creator.search(query=query, top_k=top_k, threshold=threshold)\n\n" - } -] \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/embedding_text.txt b/creator/skill_library/open-creator/search/embedding_text.txt deleted file mode 100644 index 1651ac4..0000000 --- a/creator/skill_library/open-creator/search/embedding_text.txt +++ /dev/null @@ -1,4 +0,0 @@ -search -This skill allows users to search for skills by query. -skills = search('I want to extract some pages from a pdf') -['search', 'query', 'CodeSkill'] \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/function_call.json b/creator/skill_library/open-creator/search/function_call.json deleted file mode 100644 index 3a175cd..0000000 --- a/creator/skill_library/open-creator/search/function_call.json +++ /dev/null @@ -1,26 +0,0 @@ -{ - "name": "search", - "description": "This skill allows users to search for skills by query.\n\nskills = search('I want to extract some pages from a pdf')", - "parameters": { - "type": "object", - "properties": { - "query": { - "type": "string", - "description": "The query to search for skills." - }, - "top_k": { - "type": "integer", - "description": "The maximum number of skills to return.", - "default": 1 - }, - "threshold": { - "type": "float", - "description": "The minimum similarity score to return a skill.", - "default": 0.8 - } - }, - "required": [ - "query" - ] - } -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/install_dependencies.sh b/creator/skill_library/open-creator/search/install_dependencies.sh deleted file mode 100644 index eb1ad2d..0000000 --- a/creator/skill_library/open-creator/search/install_dependencies.sh +++ /dev/null @@ -1 +0,0 @@ -pip install -U "open-creator" \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/search.py b/creator/skill_library/open-creator/search/search.py deleted file mode 100644 index 9c1a6df..0000000 --- a/creator/skill_library/open-creator/search/search.py +++ /dev/null @@ -1,22 +0,0 @@ -from creator.core import creator -from creator.core.skill import CodeSkill - - -def search(query: str, top_k=1, threshold=0.8) -> list[CodeSkill]: - """ - Search skills by query. - - Parameters: - query: str, the query. - top_k: int, optional, the maximum number of skills to return. - threshold: float, optional, the minimum similarity score to return a skill. - Returns: - a list of CodeSkill objects. - - Example: - >>> import creator - >>> skills = search("I want to extract some pages from a pdf") - """ - - return creator.search(query=query, top_k=top_k, threshold=threshold) - diff --git a/creator/skill_library/open-creator/search/skill.json b/creator/skill_library/open-creator/search/skill.json deleted file mode 100644 index 24202a7..0000000 --- a/creator/skill_library/open-creator/search/skill.json +++ /dev/null @@ -1,64 +0,0 @@ -{ - "skill_name": "search", - "skill_description": "This skill allows users to search for skills by query.", - "skill_metadata": { - "created_at": "2023-10-04 14:51:53", - "author": "gongjunmin", - "updated_at": "2023-10-04 14:51:53", - "usage_count": 0, - "version": "1.0.0", - "additional_kwargs": {} - }, - "skill_tags": [ - "search", - "query", - "CodeSkill" - ], - "skill_usage_example": "skills = search('I want to extract some pages from a pdf')", - "skill_program_language": "python", - "skill_code": "from creator.core import creator\nfrom creator.core.skill import CodeSkill\n\ndef search(query: str, top_k=1, threshold=0.8) -> list[CodeSkill]:\n '''\n Search skills by query.\n \n Parameters:\n query: str, the query.\n top_k: int, optional, the maximum number of skills to return.\n threshold: float, optional, the minimum similarity score to return a skill.\n Returns:\n a list of CodeSkill objects.\n\n Example:\n >>> import creator\n >>> skills = search('I want to extract some pages from a pdf')\n '''\n\n return creator.search(query=query, top_k=top_k, threshold=threshold)", - "skill_parameters": [ - { - "param_name": "query", - "param_type": "string", - "param_description": "The query to search for skills.", - "param_required": true, - "param_default": null - }, - { - "param_name": "top_k", - "param_type": "integer", - "param_description": "The maximum number of skills to return.", - "param_required": false, - "param_default": 1 - }, - { - "param_name": "threshold", - "param_type": "float", - "param_description": "The minimum similarity score to return a skill.", - "param_required": false, - "param_default": 0.8 - } - ], - "skill_return": { - "param_name": "skills", - "param_type": "array", - "param_description": "A list of CodeSkill objects.", - "param_required": true, - "param_default": null - }, - "skill_dependencies": [ - { - "dependency_name": "open-creator", - "dependency_version": "latest", - "dependency_type": "package" - } - ], - "conversation_history": [ - { - "role": "user", - "content": "# file name: search.py\nimport creator\nfrom creator.schema.skill import CodeSkill\n\n\ndef search(query: str, top_k=1, threshold=0.8) -> list[CodeSkill]:\n \"\"\"\n Search skills by query.\n \n Parameters:\n query: str, the query.\n top_k: int, optional, the maximum number of skills to return.\n threshold: float, optional, the minimum similarity score to return a skill.\n Returns:\n a list of CodeSkill objects.\n\n Example:\n >>> import creator\n >>> skills = search(\"I want to extract some pages from a pdf\")\n \"\"\"\n\n return creator.search(query=query, top_k=top_k, threshold=threshold)\n\n" - } - ], - "test_summary": null -} \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/skill_code.py b/creator/skill_library/open-creator/search/skill_code.py deleted file mode 100644 index c8ff216..0000000 --- a/creator/skill_library/open-creator/search/skill_code.py +++ /dev/null @@ -1,20 +0,0 @@ -import creator -from creator.core.skill import CodeSkill - -def search(query: str, top_k=1, threshold=0.8) -> list[CodeSkill]: - ''' - Search skills by query. - - Parameters: - query: str, the query. - top_k: int, optional, the maximum number of skills to return. - threshold: float, optional, the minimum similarity score to return a skill. - Returns: - a list of CodeSkill objects. - - Example: - >>> import creator - >>> skills = search('I want to extract some pages from a pdf') - ''' - - return creator.search(query=query, top_k=top_k, threshold=threshold) \ No newline at end of file diff --git a/creator/skill_library/open-creator/search/skill_doc.md b/creator/skill_library/open-creator/search/skill_doc.md deleted file mode 100644 index 9387901..0000000 --- a/creator/skill_library/open-creator/search/skill_doc.md +++ /dev/null @@ -1,18 +0,0 @@ -## Skill Details: -- **Name**: search -- **Description**: This skill allows users to search for skills by query. -- **Version**: 1.0.0 -- **Usage**: -```python -skills = search('I want to extract some pages from a pdf') -``` -- **Parameters**: - - **query** (string): The query to search for skills. - - Required: True - - **top_k** (integer): The maximum number of skills to return. - - Default: 1 - - **threshold** (float): The minimum similarity score to return a skill. - - Default: 0.8 - -- **Returns**: - - **skills** (array): A list of CodeSkill objects. \ No newline at end of file diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index 7d89f80..b1ac7a8 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -12,6 +12,8 @@ from .valid_code import is_valid_code, is_expression from .tips_utils import remove_tips from .runnable_decorator import runnable, print_run_url +from .attri_dict import AttrDict +from .uuid_generator import generate_uuid_like_string __all__ = [ @@ -30,5 +32,7 @@ "is_expression", "remove_tips", "runnable", - "print_run_url" + "print_run_url", + "AttrDict", + "generate_uuid_like_string" ] diff --git a/creator/utils/attri_dict.py b/creator/utils/attri_dict.py new file mode 100644 index 0000000..0ac26ff --- /dev/null +++ b/creator/utils/attri_dict.py @@ -0,0 +1,21 @@ +class AttrDict(dict): + def __init__(self, *args, **kwargs): + super().__init__(*args, **kwargs) + for key, value in self.items(): + if isinstance(value, dict): + self[key] = AttrDict(value) + + def __getattr__(self, key): + try: + return self[key] + except KeyError: + return None + + def __setattr__(self, key, value): + if isinstance(value, dict): + value = AttrDict(value) + self[key] = value + + def __delattr__(self, key): + if key in self: + del self[key] diff --git a/creator/utils/uuid_generator.py b/creator/utils/uuid_generator.py new file mode 100644 index 0000000..307a7ff --- /dev/null +++ b/creator/utils/uuid_generator.py @@ -0,0 +1,23 @@ +import hashlib + + +def generate_uuid_like_string(text): + """ + Generates a UUID-like string based on the given text. + + The function uses the SHA-256 hash function to hash the input text. + It then extracts 32 characters from the hash result and formats + them to mimic the structure of a UUID. + + Parameters: + text (str): The input text to be hashed and converted into a UUID-like string. + + Returns: + str: A UUID-like string derived from the input text. + """ + # Use the SHA-256 hash function to hash the input text + hash_object = hashlib.sha256(text.encode()) + hex_dig = hash_object.hexdigest() + + # Extract 32 characters from the hash result and add separators to mimic the UUID format + return f"{hex_dig[:8]}-{hex_dig[8:12]}-{hex_dig[12:16]}-{hex_dig[16:20]}-{hex_dig[20:32]}" diff --git a/pyproject.toml b/pyproject.toml index 35d5428..073df81 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -23,12 +23,12 @@ prompt_toolkit = ">=3.0.36" inquirer = "^3.1.3" pyyaml = "^6.0.1" appdirs = "^1.4.4" -urllib3 = "^2.0.6" fastapi = "^0.103.1" uvicorn = "^0.23.2" streamlit = "^1.27.2" questionary = "^2.0.1" langsmith = "^0.0.43" +qdrant-client = "^1.6.4" [tool.poetry.dependencies.pyreadline3] version = "^3.4.1" From 670e46df61308a5406954243b8fb9f46e8f820a0 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Mon, 30 Oct 2023 19:53:29 +0800 Subject: [PATCH 40/56] feat: refact create agent from config --- creator/agents/creator_agent.py | 5 +++-- creator/agents/extractor_agent.py | 9 +++++---- creator/agents/interpreter_agent.py | 5 +++-- creator/agents/prompt_enhancer_agent.py | 6 +++--- creator/agents/refactor_agent.py | 6 +++--- creator/agents/tester_agent.py | 5 +++-- creator/app/server.py | 3 +-- creator/app/streamlit_app.py | 5 ++--- creator/client/repl/handler.py | 5 ++--- creator/core/runnable.py | 17 ++++++++--------- 10 files changed, 33 insertions(+), 33 deletions(-) diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py index 6920155..5bd5a37 100644 --- a/creator/agents/creator_agent.py +++ b/creator/agents/creator_agent.py @@ -10,6 +10,7 @@ from creator.code_interpreter.safe_python import SafePythonInterpreter from creator.config.library import config from creator.utils import load_system_prompt, get_user_info, remove_tips +from creator.llm import create_llm from .base import BaseAgent @@ -66,14 +67,14 @@ async def ainvoke(self, inputs: Dict[str, Any], config: RunnableConfig | None = return {"messages": self.run(inputs)} -def create_creator_agent(llm): +def create_creator_agent(config): template = load_system_prompt(config.creator_agent_prompt_path) code_interpreter = SafePythonInterpreter(allowed_functions=ALLOWED_FUNCTIONS, allowed_methods=ALLOW_METHODS, redirect_output=True) code_interpreter.setup(IMPORT_CODE) chain = CreatorAgent( - llm=llm, + llm=create_llm(config), system_template=template, tools=[code_interpreter], function_schemas=[code_interpreter.to_function_schema()], diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py index 298bb4a..d7eb87b 100644 --- a/creator/agents/extractor_agent.py +++ b/creator/agents/extractor_agent.py @@ -1,10 +1,11 @@ from typing import Dict, Any +import json + from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json -from creator.config.library import config from creator.utils import convert_to_values_list, get_user_info, load_system_prompt -import json +from creator.llm import create_llm from .base import BaseAgent @@ -35,7 +36,7 @@ def parse_output(self, messages): return {"extracted_skill": None} -def create_skill_extractor_agent(llm): +def create_skill_extractor_agent(config): template = load_system_prompt(config.extractor_agent_prompt_path) # current file's parent as dir with open(config.codeskill_function_schema_path, encoding="utf-8") as f: @@ -47,7 +48,7 @@ def create_skill_extractor_agent(llm): } chain = SkillExtractorAgent( - llm=llm, + llm=create_llm(config), system_template=template, function_schemas=[function_schema], verbose=False diff --git a/creator/agents/interpreter_agent.py b/creator/agents/interpreter_agent.py index 71407eb..68c63c7 100644 --- a/creator/agents/interpreter_agent.py +++ b/creator/agents/interpreter_agent.py @@ -6,6 +6,7 @@ from creator.code_interpreter import CodeInterpreter, language_map from creator.config.library import config from creator.utils import load_system_prompt, remove_tips +from creator.llm import create_llm from .base import BaseAgent @@ -48,12 +49,12 @@ def messages_hot_fix(self, langchain_messages): return langchain_messages -def create_code_interpreter_agent(llm): +def create_code_interpreter_agent(config): tool = CodeInterpreter() function_schema = tool.to_function_schema() template = load_system_prompt(config.interpreter_agent_prompt_path) chain = CodeInterpreterAgent( - llm=llm, + llm=create_llm(config), system_template=template, function_schemas=[function_schema], tools=[tool], diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py index c8425ba..35f873f 100644 --- a/creator/agents/prompt_enhancer_agent.py +++ b/creator/agents/prompt_enhancer_agent.py @@ -3,8 +3,8 @@ from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json -from creator.config.library import config from creator.utils import load_system_prompt +from creator.llm import create_llm from creator.utils import print from creator.agents.base import BaseAgent @@ -37,14 +37,14 @@ def parse_output(self, messages): return {"request": orignal_request} -def create_prompt_enhancer_agent(llm): +def create_prompt_enhancer_agent(config): template = load_system_prompt(config.prompt_enhancer_agent_prompt_path) with open(config.prompt_enhancer_schema_path, encoding="utf-8") as f: function_schema = json.load(f) chain = PromptEnhancerAgent( - llm=llm, + llm=create_llm(config), system_template=template, function_schemas=[function_schema], verbose=False diff --git a/creator/agents/refactor_agent.py b/creator/agents/refactor_agent.py index 6c69940..51600ed 100644 --- a/creator/agents/refactor_agent.py +++ b/creator/agents/refactor_agent.py @@ -5,7 +5,7 @@ from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json -from creator.config.library import config +from creator.llm import create_llm from creator.utils import convert_to_values_list, load_system_prompt, get_user_info from .base import BaseAgent @@ -38,7 +38,7 @@ def parse_output(self, messages): return {"refacted_skills": None} -def create_code_refactor_agent(llm): +def create_code_refactor_agent(config): template = load_system_prompt(config.refactor_agent_prompt_path) path = os.path.join(config.codeskill_function_schema_path) with open(path, encoding="utf-8") as f: @@ -61,7 +61,7 @@ def create_code_refactor_agent(llm): } chain = CodeRefactorAgent( - llm=llm, + llm=create_llm(config), system_template=template, function_schemas=[function_schema], verbose=False diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py index ae6e19a..a7c4971 100644 --- a/creator/agents/tester_agent.py +++ b/creator/agents/tester_agent.py @@ -5,6 +5,7 @@ from creator.code_interpreter import CodeInterpreter, language_map from creator.config.library import config +from creator.llm import create_llm from creator.utils import load_system_prompt, remove_tips from .base import BaseAgent @@ -65,7 +66,7 @@ def parse_output(self, messages): } -def create_code_tester_agent(llm): +def create_code_tester_agent(config): template = load_system_prompt(config.tester_agent_prompt_path) tool = CodeInterpreter() @@ -74,7 +75,7 @@ def create_code_tester_agent(llm): test_summary_function_schema = json.load(f) chain = CodeTesterAgent( - llm=llm, + llm=create_llm(config), system_template=template, function_schemas=[code_interpreter_function_schema, test_summary_function_schema], tools=[tool], diff --git a/creator/app/server.py b/creator/app/server.py index 511a0e7..9dd2c5c 100644 --- a/creator/app/server.py +++ b/creator/app/server.py @@ -1,6 +1,5 @@ from fastapi import FastAPI from creator.agents.creator_agent import create_creator_agent -from creator.llm import create_llm from creator import config from creator.__version__ import __version__ as version from pydantic import BaseModel @@ -21,7 +20,7 @@ class Output(BaseModel): ) config.use_rich = False -open_creator_agent = create_creator_agent(create_llm(config)) +open_creator_agent = create_creator_agent(config) @app.post("/agents/creator") diff --git a/creator/app/streamlit_app.py b/creator/app/streamlit_app.py index bbb6870..d41fdc9 100644 --- a/creator/app/streamlit_app.py +++ b/creator/app/streamlit_app.py @@ -5,7 +5,6 @@ import streamlit as st from creator.agents import create_creator_agent, create_code_interpreter_agent -from creator.llm import create_llm from creator import config from langchain.callbacks.streamlit.streamlit_callback_handler import _convert_newlines from langchain.output_parsers.json import parse_partial_json @@ -22,8 +21,8 @@ agent_name = "interpreter_agent" agent_mapping = { - "interpreter_agent": create_code_interpreter_agent(create_llm(config)), - "creator_agent": create_creator_agent(create_llm(config)) + "interpreter_agent": create_code_interpreter_agent(config), + "creator_agent": create_creator_agent(config) } diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py index b6f9238..e91be5f 100644 --- a/creator/client/repl/handler.py +++ b/creator/client/repl/handler.py @@ -4,7 +4,6 @@ from creator.agents import create_creator_agent, create_code_interpreter_agent from creator.config.library import config -from creator.llm import create_llm from creator.utils import truncate_output, is_valid_code from rich.console import Console from rich.markdown import Markdown @@ -22,8 +21,8 @@ def __init__(self): self.output = [] self.console = Console() self.interpreter = False - self.interpreter_agent = create_code_interpreter_agent(create_llm(config)) - self.open_creator_agent = create_creator_agent(create_llm(config)) + self.interpreter_agent = create_code_interpreter_agent(config) + self.open_creator_agent = create_creator_agent(config) async def handle(self, request, interpreter): """ diff --git a/creator/core/runnable.py b/creator/core/runnable.py index 1dd152f..da64ad6 100644 --- a/creator/core/runnable.py +++ b/creator/core/runnable.py @@ -5,7 +5,6 @@ from creator.utils import runnable, print, print_run_url from creator.utils import generate_install_command from creator.config.library import config as creator_config -from creator.llm import create_llm from creator.agents import ( create_skill_extractor_agent, create_code_interpreter_agent, @@ -30,17 +29,17 @@ def construct_create_skill_messages(request): @print_run_url def create_skill_from_messages(messages): - skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) + skill_extractor_agent = create_skill_extractor_agent(creator_config) return skill_extractor_agent.with_config({"run_name": "CreateSkillFromMessages"}).invoke(input={"messages": messages})["extracted_skill"] @print_run_url def create_skill_from_request(request): creator_config.use_rich = False - prompt_enhancer_agent = create_prompt_enhancer_agent(create_llm(creator_config)) + prompt_enhancer_agent = create_prompt_enhancer_agent(creator_config) creator_config.use_rich = True - code_interpreter_agent = create_code_interpreter_agent(create_llm(creator_config)) - skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) + code_interpreter_agent = create_code_interpreter_agent(creator_config) + skill_extractor_agent = create_skill_extractor_agent(creator_config) chain = construct_create_skill_messages | prompt_enhancer_agent | construct_create_skill_messages | code_interpreter_agent | skill_extractor_agent skill_json = chain.with_config({"run_name": "CreateSkillFromRequest"}).invoke(input=request)["extracted_skill"] return skill_json @@ -48,7 +47,7 @@ def create_skill_from_request(request): @print_run_url def create_skill_from_file_content(file_content): - skill_extractor_agent = create_skill_extractor_agent(create_llm(creator_config)) + skill_extractor_agent = create_skill_extractor_agent(creator_config) chain = construct_create_skill_messages | skill_extractor_agent skill_json = chain.with_config({"run_name": "CreateSkillFromFileContent"}).invoke(input=file_content)["extracted_skill"] return skill_json @@ -101,7 +100,7 @@ def setup_skill(inputs, config: RunnableConfig): @runnable(run_name="RunSkill") def run_skill(inputs, config: RunnableConfig): - code_interpreter_agent = create_code_interpreter_agent(create_llm(config=creator_config)) + code_interpreter_agent = create_code_interpreter_agent(creator_config) code_interpreter_agent.tools[0] = creator_config.code_interpreter tool_inputs = {"language": inputs["skill_program_language"], "code": inputs["skill_code"]} inputs.update(tool_inputs) @@ -125,7 +124,7 @@ def construct_test_skill_messages(inputs): @print_run_url @runnable(run_name="TestSkill") def test_skill(inputs, config: RunnableConfig): - code_tester_agent = create_code_tester_agent(create_llm(creator_config)) + code_tester_agent = create_code_tester_agent(creator_config) code_tester_agent.tools[0] = creator_config.code_interpreter code = f"""\n\n import io @@ -160,7 +159,7 @@ def construct_refactor_skill_messages(inputs): @print_run_url @runnable(run_name="RefactorSkill") def refactor_skill(inputs, config: RunnableConfig): - code_refactor_agent = create_code_refactor_agent(create_llm(creator_config)) + code_refactor_agent = create_code_refactor_agent(creator_config) chain = construct_refactor_skill_messages | code_refactor_agent refactored_skill_jsons = chain.invoke(inputs, config)["refacted_skills"] return refactored_skill_jsons From 0bc3c7f829963b02c2ff7eb8d46119f76c3cd7aa Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 31 Oct 2023 11:09:45 +0800 Subject: [PATCH 41/56] feat: use config to directly create agents and support load function schema utils --- creator/agents/creator_agent.py | 2 -- creator/agents/extractor_agent.py | 6 ++---- creator/agents/prompt_enhancer_agent.py | 8 ++------ creator/agents/refactor_agent.py | 7 ++----- creator/agents/tester_agent.py | 6 ++---- creator/memgpt/agent.py | 26 ++++++++++++------------- creator/memgpt/memory/__init__.py | 16 +++++++-------- creator/utils/__init__.py | 3 ++- creator/utils/load_prompt.py | 9 +++++++++ 9 files changed, 39 insertions(+), 44 deletions(-) diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py index 5bd5a37..c2167f7 100644 --- a/creator/agents/creator_agent.py +++ b/creator/agents/creator_agent.py @@ -69,10 +69,8 @@ async def ainvoke(self, inputs: Dict[str, Any], config: RunnableConfig | None = def create_creator_agent(config): template = load_system_prompt(config.creator_agent_prompt_path) - code_interpreter = SafePythonInterpreter(allowed_functions=ALLOWED_FUNCTIONS, allowed_methods=ALLOW_METHODS, redirect_output=True) code_interpreter.setup(IMPORT_CODE) - chain = CreatorAgent( llm=create_llm(config), system_template=template, diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py index d7eb87b..21a2f6f 100644 --- a/creator/agents/extractor_agent.py +++ b/creator/agents/extractor_agent.py @@ -1,10 +1,9 @@ from typing import Dict, Any -import json from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json -from creator.utils import convert_to_values_list, get_user_info, load_system_prompt +from creator.utils import convert_to_values_list, get_user_info, load_system_prompt, load_json_schema from creator.llm import create_llm from .base import BaseAgent @@ -39,8 +38,7 @@ def parse_output(self, messages): def create_skill_extractor_agent(config): template = load_system_prompt(config.extractor_agent_prompt_path) # current file's parent as dir - with open(config.codeskill_function_schema_path, encoding="utf-8") as f: - code_skill_json_schema = json.load(f) + code_skill_json_schema = load_json_schema(config.codeskill_function_schema_path) function_schema = { "name": "extract_formmated_skill", "description": "a function that extracts a skill from a conversation history", diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py index 35f873f..1434d2d 100644 --- a/creator/agents/prompt_enhancer_agent.py +++ b/creator/agents/prompt_enhancer_agent.py @@ -3,12 +3,11 @@ from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json -from creator.utils import load_system_prompt +from creator.utils import load_system_prompt, load_json_schema from creator.llm import create_llm from creator.utils import print from creator.agents.base import BaseAgent -import json class PromptEnhancerAgent(BaseAgent): @@ -39,10 +38,7 @@ def parse_output(self, messages): def create_prompt_enhancer_agent(config): template = load_system_prompt(config.prompt_enhancer_agent_prompt_path) - - with open(config.prompt_enhancer_schema_path, encoding="utf-8") as f: - function_schema = json.load(f) - + function_schema = load_json_schema(config.prompt_enhancer_schema_path) chain = PromptEnhancerAgent( llm=create_llm(config), system_template=template, diff --git a/creator/agents/refactor_agent.py b/creator/agents/refactor_agent.py index 51600ed..7b0b479 100644 --- a/creator/agents/refactor_agent.py +++ b/creator/agents/refactor_agent.py @@ -1,12 +1,11 @@ from typing import Any, Dict -import json import os from langchain.prompts import ChatPromptTemplate from langchain.output_parsers.json import parse_partial_json from creator.llm import create_llm -from creator.utils import convert_to_values_list, load_system_prompt, get_user_info +from creator.utils import convert_to_values_list, load_system_prompt, get_user_info, load_json_schema from .base import BaseAgent @@ -41,9 +40,7 @@ def parse_output(self, messages): def create_code_refactor_agent(config): template = load_system_prompt(config.refactor_agent_prompt_path) path = os.path.join(config.codeskill_function_schema_path) - with open(path, encoding="utf-8") as f: - code_skill_json_schema = json.load(f) - + code_skill_json_schema = load_json_schema(os.path.join(config.codeskill_function_schema_path)) function_schema = { "name": "create_refactored_codeskills", "description": "a function that constructs a list of refactored skill objects. return only one item when your action is to combine or refine skill object(s), otherwise return more than one items", diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py index a7c4971..3e8d194 100644 --- a/creator/agents/tester_agent.py +++ b/creator/agents/tester_agent.py @@ -6,7 +6,7 @@ from creator.code_interpreter import CodeInterpreter, language_map from creator.config.library import config from creator.llm import create_llm -from creator.utils import load_system_prompt, remove_tips +from creator.utils import load_system_prompt, load_json_schema, remove_tips from .base import BaseAgent @@ -71,9 +71,7 @@ def create_code_tester_agent(config): tool = CodeInterpreter() code_interpreter_function_schema = tool.to_function_schema() - with open(config.testsummary_function_schema_path, encoding="utf-8") as f: - test_summary_function_schema = json.load(f) - + test_summary_function_schema = load_json_schema(config.testsummary_function_schema_path) chain = CodeTesterAgent( llm=create_llm(config), system_template=template, diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py index 276cb55..a41e180 100644 --- a/creator/memgpt/agent.py +++ b/creator/memgpt/agent.py @@ -1,6 +1,5 @@ from typing import List, Dict, Any, Optional import datetime -import json from langchain.schema.messages import FunctionMessage from langchain.output_parsers.json import parse_partial_json @@ -8,16 +7,16 @@ from langchain.prompts import ChatPromptTemplate from langchain.adapters.openai import convert_openai_messages -from creator.utils import load_system_prompt +from creator.utils import load_system_prompt, load_json_schema from creator.agents.base import BaseAgent +from creator.agents import create_creator_agent +from creator.llm import create_llm from .memory import MemoryManager from .constants import ( - DEFAULT_AGENT_SUBTASKS, MESSAGE_SUMMARY_WARNING_STR, FUNC_FAILED_HEARTBEAT_MESSAGE, - REQ_HEARTBEAT_MESSAGE, - MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE + REQ_HEARTBEAT_MESSAGE ) from .message_handler import ( get_initial_boot_messages, @@ -59,8 +58,7 @@ def preprocess_inputs(self, inputs: Dict[str, Any]): inputs["persona"] = self.memory_manager.persona inputs["human"] = self.memory_manager.human - if "subagent_tasks" not in inputs: - inputs["subagent_tasks"] = DEFAULT_AGENT_SUBTASKS + inputs["subagent_tasks"] = self.memgpt_config.AGENT_SUBTASKS return inputs @@ -210,20 +208,20 @@ async def _acall( return output -def create_memgpt(llm, memgpt_config, subagent=None): - template = load_system_prompt(memgpt_config.system_prompt_path) - with open(memgpt_config.function_schema_path) as f: - function_schemas = json.load(f) - memory_manager = MemoryManager(memgpt_config) +def create_memgpt(config, subagent=None): + template = load_system_prompt(config.memgpt_system_prompt_path) + function_schemas = load_json_schema(config.memgpt__function_schema_path) + memory_manager = MemoryManager(config.memgpt_config) + llm = create_llm(config) if subagent is None: - subagent = BaseAgent(llm=llm, system_template=MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE) + subagent = create_creator_agent(config) chain = MemGPT( llm=llm, subagent=subagent, system_template=template, function_schemas=function_schemas, memory_manager=memory_manager, - memgpt_config=memgpt_config, + memgpt_config=config.memgpt_config, output_key="session_id", verbose=False, ) diff --git a/creator/memgpt/memory/__init__.py b/creator/memgpt/memory/__init__.py index 2c1e571..6cf1953 100644 --- a/creator/memgpt/memory/__init__.py +++ b/creator/memgpt/memory/__init__.py @@ -8,17 +8,17 @@ class MemoryManager: def __init__(self, memory_config): - self.chat_message_history = build_memory(memory_path=memory_config.memory_path, session_id=memory_config.session_id) + self.chat_message_history = build_memory(memory_path=memory_config.MEMORY_PATH, session_id=memory_config.session_id) self.session_id = self.chat_message_history.session_id self.core_memory = CoreMemory( - persona=memory_config.persona, - human=memory_config.human, - persona_char_limit=memory_config.persona_char_limit, - human_char_limit=memory_config.human_char_limit, + persona=memory_config.PERSONA, + human=memory_config.HUMAN, + persona_char_limit=memory_config.CORE_MEMORY_PERSONA_CHAR_LIMIT, + human_char_limit=memory_config.CORE_MEMORY_HUMAN_CHAR_LIMIT, ) - self.recall_memory = RecallMemory(message_database=self.chat_message_history, use_vector_search=memory_config.use_vector_search) - self.archival_memory = ArchivalMemory(message_database=self.chat_message_history, use_vector_search=memory_config.use_vector_search) - self.page_size = self.recall_memory.page_size = self.archival_memory.page_size = memory_config.page_size + self.recall_memory = RecallMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH) + self.archival_memory = ArchivalMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH) + self.page_size = self.recall_memory.page_size = self.archival_memory.page_size = memory_config.PAGE_SIZE @property def human(self): diff --git a/creator/utils/__init__.py b/creator/utils/__init__.py index b1ac7a8..15cb188 100644 --- a/creator/utils/__init__.py +++ b/creator/utils/__init__.py @@ -6,7 +6,7 @@ from .ask_human import ask_run_code_confirm from .dict2list import convert_to_values_list from .user_info import get_user_info -from .load_prompt import load_system_prompt +from .load_prompt import load_system_prompt, load_json_schema from .printer import print from .code_split import split_code_blocks from .valid_code import is_valid_code, is_expression @@ -26,6 +26,7 @@ "convert_to_values_list", "get_user_info", "load_system_prompt", + "load_json_schema", "print", "split_code_blocks", "is_valid_code", diff --git a/creator/utils/load_prompt.py b/creator/utils/load_prompt.py index 00bccde..c73cb89 100644 --- a/creator/utils/load_prompt.py +++ b/creator/utils/load_prompt.py @@ -1,4 +1,13 @@ +import json + + def load_system_prompt(prompt_path): with open(prompt_path, encoding='utf-8') as f: prompt = f.read() return prompt + + +def load_json_schema(json_schema_path): + with open(json_schema_path, encoding="utf-8") as f: + json_schema = json.load(f) + return json_schema From 35c4d15d09f1926bc25fb19f6236924a45c05714 Mon Sep 17 00:00:00 2001 From: Gong Junmin <1836678486@qq.com> Date: Tue, 31 Oct 2023 14:19:32 +0800 Subject: [PATCH 42/56] fix: better function schema support for openai --- creator/agents/base.py | 3 - creator/agents/extractor_agent.py | 13 +- creator/llm/format_function_calls.py | 8 +- .../prompts/codeskill_function_schema.json | 84 +---- .../prompts/testsummary_function_schema.json | 67 ++-- creator/utils/runnable_decorator.py | 2 +- docs/examples/01_skills_create.ipynb | 326 ++++++++++++------ 7 files changed, 281 insertions(+), 222 deletions(-) diff --git a/creator/agents/base.py b/creator/agents/base.py index 3aacbe5..3a0f186 100644 --- a/creator/agents/base.py +++ b/creator/agents/base.py @@ -85,9 +85,6 @@ def run_tool(self, function_call: Dict[str, Any], run_manager: Optional[Callback tool_result = FunctionMessage(name=function_name, content=tool_result) self.update_tool_result_in_callbacks(tool_result) break - if tool_result is None: - tool_names = list(map(lambda x: x.name, self.tools)) - tool_result = FunctionMessage(name="system_alert", content=f"Unknown function name: {function_name}. Available functions: {tool_names}") return tool_result def human_confirm(self): diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py index 21a2f6f..1d8b87d 100644 --- a/creator/agents/extractor_agent.py +++ b/creator/agents/extractor_agent.py @@ -22,16 +22,21 @@ def construct_prompt(self, langchain_messages: Dict[str, Any]): ("system", self.system_template + get_user_info()) ]) return prompt - + def parse_output(self, messages): function_call = messages[-1].get("function_call", None) - - if function_call is not None: - extracted_skill = parse_partial_json(function_call.get("arguments", "{}")) + try: + if function_call is not None: + content = function_call.get("arguments", "{}") + else: + content = messages[-1].get("content", "{}") + extracted_skill = parse_partial_json(content) extracted_skill["conversation_history"] = messages[:-1] extracted_skill["skill_parameters"] = convert_to_values_list(extracted_skill["skill_parameters"]) if "skill_parameters" in extracted_skill else None extracted_skill["skill_return"] = convert_to_values_list(extracted_skill["skill_return"]) if "skill_return" in extracted_skill else None return {"extracted_skill": extracted_skill} + except Exception: + pass return {"extracted_skill": None} diff --git a/creator/llm/format_function_calls.py b/creator/llm/format_function_calls.py index 7f0906b..7d326be 100644 --- a/creator/llm/format_function_calls.py +++ b/creator/llm/format_function_calls.py @@ -19,8 +19,10 @@ def resolve_ref(schema, json_schema): if schema.get("$ref") is not None: - ref = schema["$ref"][14:] - schema = json_schema["definitions"][ref] + # #/$defs/CodeSkillParameter + ref = schema["$ref"][8:] + if "$defs" in json_schema and ref in json_schema["$defs"]: + return json_schema["$defs"][ref] return schema @@ -63,6 +65,8 @@ def format_object(schema, indent, json_schema): def format_schema(schema, indent, json_schema): schema = resolve_ref(schema, json_schema) + if "type" not in schema: + return "any" if "enum" in schema: return format_enum(schema) elif schema["type"] == "object": diff --git a/creator/prompts/codeskill_function_schema.json b/creator/prompts/codeskill_function_schema.json index 93769ce..c462ddf 100644 --- a/creator/prompts/codeskill_function_schema.json +++ b/creator/prompts/codeskill_function_schema.json @@ -33,57 +33,26 @@ "type": "string" }, "skill_parameters": { - "anyOf": [ - { - "$ref": "#/$defs/CodeSkillParameter" - }, - { - "items": { - "$ref": "#/$defs/CodeSkillParameter" - }, - "type": "array" - }, - { - "type": "null" - } - ], + "items": { + "$ref": "#/$defs/CodeSkillParameter" + }, + "type": "array", "default": null, "description": "List of parameters the skill requires, defined using json schema" }, "skill_return": { - "anyOf": [ - { + "items": { "$ref": "#/$defs/CodeSkillParameter" }, - { - "items": { - "$ref": "#/$defs/CodeSkillParameter" - }, - "type": "array" - }, - { - "type": "null" - } - ], + "type": "array", "default": null, "description": "Return value(s) of the skill" }, "skill_dependencies": { - "anyOf": [ - { + "items": { "$ref": "#/$defs/CodeSkillDependency" }, - { - "items": { - "$ref": "#/$defs/CodeSkillDependency" - }, - "type": "array" - }, - { - "type": "null" - } - ], - "default": null, + "type": "array", "description": "List of dependencies the skill requires to run, typically packages but can also be other skill functions" }, "skill_usage_example": { @@ -94,7 +63,6 @@ "required": [ "skill_name", "skill_tags", - "skill_usage_example", "skill_program_language", "skill_code" ], @@ -103,30 +71,14 @@ "CodeSkillDependency": { "properties": { "dependency_name": { - "default": "", "type": "string" }, "dependency_version": { - "anyOf": [ - { - "type": "string" - }, - { - "type": "null" - } - ], - "default": "", + "type": "string", "description": "the version of the dependency only filled if context provided" }, "dependency_type": { - "anyOf": [ - { - "type": "string" - }, - { - "type": "null" - } - ], + "type": "string", "default": "built-in", "description": "when the dependency is an another code skill, please set it as function", "enum": [ @@ -136,17 +88,16 @@ ] } }, - "type": "object" + "type": "object", + "required": ["dependency_name", "dependency_type"] }, "CodeSkillParameter": { "properties": { "param_name": { - "default": "query", "description": "the name of the parameter", "type": "string" }, "param_type": { - "default": "string", "description": "the type, only support string, integer, float, boolean, array, object", "enum": [ "string", @@ -159,7 +110,6 @@ "type": "string" }, "param_description": { - "default": "the input query", "description": "the description of the parameter. If it is enum, describe the enum values. If it is format, describe the format", "type": "string" }, @@ -169,17 +119,11 @@ "type": "boolean" }, "param_default": { - "anyOf": [ - {}, - { - "type": "null" - } - ], - "default": null, "description": "the default value, it depends on the type" } }, - "type": "object" + "type": "object", + "required": ["param_name", "param_type", "param_required"] } } } \ No newline at end of file diff --git a/creator/prompts/testsummary_function_schema.json b/creator/prompts/testsummary_function_schema.json index 2343e9c..5fa0648 100644 --- a/creator/prompts/testsummary_function_schema.json +++ b/creator/prompts/testsummary_function_schema.json @@ -2,45 +2,40 @@ "name": "test_summary", "description": "A method to be invoked once all test cases have been successfully completed. This function provides a comprehensive summary of each test case, detailing their input, execution command, expected results, actual results, and pass status.", "parameters": { - "$defs": { - "TestCase": { - "properties": { - "test_input": { - "description": "The input data or conditions used for the test.", - "type": "string" - }, - "run_command": { - "description": "The command or function that was executed for the test.", - "type": "string" - }, - "expected_result": { - "description": "The expected outcome or result of the test.", - "type": "string" - }, - "actual_result": { - "description": "The actual outcome or result observed after the test was executed.", - "type": "string" - }, - "is_passed": { - "description": "A boolean indicating whether the test passed or failed.", - "type": "boolean" - } - }, - "required": [ - "test_input", - "run_command", - "expected_result", - "actual_result", - "is_passed" - ], - "type": "object" - } - }, "properties": { "test_cases": { "description": "Extract a list of test cases that were run.", "items": { - "$ref": "#/$defs/TestCase" + "properties": { + "test_input": { + "description": "The input data or conditions used for the test.", + "type": "string" + }, + "run_command": { + "description": "The command or function that was executed for the test.", + "type": "string" + }, + "expected_result": { + "description": "The expected outcome or result of the test.", + "type": "string" + }, + "actual_result": { + "description": "The actual outcome or result observed after the test was executed.", + "type": "string" + }, + "is_passed": { + "description": "A boolean indicating whether the test passed or failed.", + "type": "boolean" + } + }, + "required": [ + "test_input", + "run_command", + "expected_result", + "actual_result", + "is_passed" + ], + "type": "object" }, "type": "array" } @@ -50,4 +45,4 @@ ], "type": "object" } -} \ No newline at end of file +} diff --git a/creator/utils/runnable_decorator.py b/creator/utils/runnable_decorator.py index d4afc23..52342c2 100644 --- a/creator/utils/runnable_decorator.py +++ b/creator/utils/runnable_decorator.py @@ -17,7 +17,7 @@ def wrapper(*args, **kwargs): result = func(*args, **kwargs) run_url = cb.get_run_url() if run_url is not None: - print(f"Langsmith Run URL: [{run_url}]{run_url}", print_type="markdown") + print(f"Langsmith Run URL: [{run_url}]({run_url})", print_type="markdown") else: result = func(*args, **kwargs) return result diff --git a/docs/examples/01_skills_create.ipynb b/docs/examples/01_skills_create.ipynb index a1acc44..e82b1be 100644 --- a/docs/examples/01_skills_create.ipynb +++ b/docs/examples/01_skills_create.ipynb @@ -62,6 +62,7 @@ "\u001b[0;34m\u001b[0m \u001b[0mfile_path\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mOptional\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mstr\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\n", "\u001b[0;34m\u001b[0m \u001b[0mhuggingface_repo_id\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mOptional\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mstr\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\n", "\u001b[0;34m\u001b[0m \u001b[0mhuggingface_skill_path\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mOptional\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mstr\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\n", + "\u001b[0;34m\u001b[0m \u001b[0msave\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mbool\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\n", "\u001b[0;34m\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m->\u001b[0m \u001b[0mcreator\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mskill\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mCodeSkill\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n", "\u001b[0;31mDocstring:\u001b[0m Main method to create a new skill.\n", "\u001b[0;31mFile:\u001b[0m ~/miniconda3/envs/open_creator_online/lib/python3.10/site-packages/creator/core/core.py\n", @@ -93,7 +94,7 @@ { "data": { "application/vnd.jupyter.widget-view+json": { - "model_id": "480c4325950f46d9a9802e3952ef5ddb", + "model_id": "59eeaa5d83ec4cf9bdea0dbead71929c", "version_major": 2, "version_minor": 0 }, @@ -107,7 +108,7 @@ { "data": { "application/vnd.jupyter.widget-view+json": { - "model_id": "00c9080c652f4b26b163fd80933c8bf3", + "model_id": "b49892b54f2547d2a74ae02760d79ab0", "version_major": 2, "version_minor": 0 }, @@ -141,7 +142,55 @@ { "data": { "application/vnd.jupyter.widget-view+json": { - "model_id": "74f96ccdc008481d92dd54ad13aecde2", + "model_id": "c1fc6b0e407b48fd9709e1fb878a6b3a", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "Output()" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "2dfb101217d84fff944e06c72db7eee7", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "Output()" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
\n"
+      ],
+      "text/plain": []
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
\n"
+      ],
+      "text/plain": []
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "bdd53aed57ca41e7a6b945d3e17f4908",
        "version_major": 2,
        "version_minor": 0
       },
@@ -175,7 +224,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "3792eb41434a4afab72082018131f2e9",
+       "model_id": "8978e8e0ba4b4955a0b97b3ba2a549e0",
        "version_major": 2,
        "version_minor": 0
       },
@@ -205,6 +254,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c2491d5f-\n",
+       "20de-4067-af80-134cafb3a449?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=641371;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c2491d5f-20de-4067-af80-134cafb3a449?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c2491d5f-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=641371;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c2491d5f-20de-4067-af80-134cafb3a449?poll=true\u001b\\\u001b[4;34m20de-4067-af80-134cafb3a449?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -223,21 +289,20 @@
        "                                                  Skill Details:                                                   \n",
        "\n",
        " • Name: count_prime_numbers                                                                                       \n",
-       " • Description: This skill counts the number of prime numbers in a given range.                                    \n",
+       " • Description: This skill counts the number of prime numbers within a given range. A prime number is a natural    \n",
+       "   number greater than 1 that has no positive divisors other than 1 and itself.                                    \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " count_prime_numbers(2, 201)                                                                                       \n",
+       " count_prime_numbers(201)  # returns 46                                                                            \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
-       "    • start (integer): The starting number of the range.                                                           \n",
-       "       • Required: True                                                                                            \n",
-       "    • end (integer): The ending number of the range.                                                               \n",
+       "    • n (integer): The upper limit of the range within which to count prime numbers.                               \n",
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
-       "    • count (integer): The number of prime numbers in the given range.                                             \n",
+       "    • prime_count (integer): The number of prime numbers within the given range.                                   \n",
        "\n"
       ],
       "text/plain": [
@@ -245,21 +310,20 @@
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: count_prime_numbers                                                                                       \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill counts the number of prime numbers in a given range.                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill counts the number of prime numbers within a given range. A prime number is a natural    \n",
+       "\u001b[1;33m   \u001b[0mnumber greater than 1 that has no positive divisors other than 1 and itself.                                    \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcount_prime_numbers\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m201\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                                      \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcount_prime_numbers\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m201\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m  \u001b[0m\u001b[38;2;149;144;119;48;2;39;40;34m# returns 46\u001b[0m\u001b[48;2;39;40;34m                                                                           \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mstart\u001b[0m (integer): The starting number of the range.                                                           \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mend\u001b[0m (integer): The ending number of the range.                                                               \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mn\u001b[0m (integer): The upper limit of the range within which to count prime numbers.                               \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mcount\u001b[0m (integer): The number of prime numbers in the given range.                                             \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mprime_count\u001b[0m (integer): The number of prime numbers within the given range.                                   \n"
       ]
      },
      "metadata": {},
@@ -310,7 +374,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "1409e75812d1493a8885087f03ac6f25",
+       "model_id": "91d14a513cff4b68ac9cba04dc1a6db4",
        "version_major": 2,
        "version_minor": 0
       },
@@ -340,6 +404,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/1e9ecd99-\n",
+       "be29-4eee-8b21-449f279b7f4b?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=254594;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/1e9ecd99-be29-4eee-8b21-449f279b7f4b?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/1e9ecd99-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=254594;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/1e9ecd99-be29-4eee-8b21-449f279b7f4b?poll=true\u001b\\\u001b[4;34mbe29-4eee-8b21-449f279b7f4b?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -357,17 +438,19 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: list_python_functions                                                                                     \n",
-       " • Description: This skill lists all the Python functions and their docstrings in a specified directory.           \n",
+       " • Name: list_python_functions_and_docstrings                                                                      \n",
+       " • Description: This skill lists all the Python functions and their docstrings in a specified directory. It first  \n",
+       "   gets a list of all Python files in the directory, then parses each file to find all function definitions. For   \n",
+       "   each function definition, it extracts the function's name and its docstring.                                    \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " list_python_functions('/path/to/directory')                                                                       \n",
+       " list_python_functions_and_docstrings('/path/to/directory')                                                        \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
-       "    • directory (string): The directory path where the Python files are located.                                   \n",
+       "    • directory (string): The directory to search for Python files.                                                \n",
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
        "
\n"
@@ -376,17 +459,19 @@
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: list_python_functions                                                                                     \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill lists all the Python functions and their docstrings in a specified directory.           \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: list_python_functions_and_docstrings                                                                      \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill lists all the Python functions and their docstrings in a specified directory. It first  \n",
+       "\u001b[1;33m   \u001b[0mgets a list of all Python files in the directory, then parses each file to find all function definitions. For   \n",
+       "\u001b[1;33m   \u001b[0meach function definition, it extracts the function's name and its docstring.                                    \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mlist_python_functions\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m/path/to/directory\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                      \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mlist_python_functions_and_docstrings\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m/path/to/directory\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mdirectory\u001b[0m (string): The directory path where the Python files are located.                                   \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mdirectory\u001b[0m (string): The directory to search for Python files.                                                \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n"
       ]
@@ -415,7 +500,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "ef6cc0331df844629f215b0e657c58e9",
+       "model_id": "72229d73f3cc4980bb7bcac6009fe5b9",
        "version_major": 2,
        "version_minor": 0
       },
@@ -445,6 +530,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c94c0ec7-\n",
+       "ce23-44f5-91d5-6bd5ec1f1e39?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=672460;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c94c0ec7-ce23-44f5-91d5-6bd5ec1f1e39?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c94c0ec7-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=672460;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/c94c0ec7-ce23-44f5-91d5-6bd5ec1f1e39?poll=true\u001b\\\u001b[4;34mce23-44f5-91d5-6bd5ec1f1e39?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -462,62 +564,56 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: extract_pdf_pages                                                                                         \n",
-       " • Description: This skill extracts a specified section from a PDF file and saves it as a new PDF.                 \n",
+       " • Name: extract_pages_from_pdf                                                                                    \n",
+       " • Description: This skill extracts a specified range of pages from a PDF file and saves them as a new PDF file.   \n",
+       "   The user needs to provide the path to the original PDF file and the range of pages to be extracted. The         \n",
+       "   extracted pages are saved in a new PDF file in the current working directory.                                   \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " pdf_path = '~/Downloads/voyager.pdf'                                                                              \n",
-       " start_page = 2                                                                                                    \n",
-       " end_page = 5                                                                                                      \n",
-       " output_path = 'extracted_pages.pdf'                                                                               \n",
-       "                                                                                                                   \n",
-       " extract_pdf_pages(pdf_path, start_page, end_page, output_path)                                                    \n",
+       " extract_pages_from_pdf('~/Downloads/voyager.pdf', 2, 5, 'voyager_extracted.pdf')                                  \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
-       "    • pdf_path (string): The path to the PDF file.                                                                 \n",
-       "       • Required: True                                                                                            \n",
-       "    • start_page (integer): The starting page number to extract.                                                   \n",
+       "    • pdf_path (string): Path to the original PDF file.                                                            \n",
        "       • Required: True                                                                                            \n",
-       "    • end_page (integer): The ending page number to extract.                                                       \n",
+       "    • start_page (integer): The first page to be extracted. Page numbers start from 1.                             \n",
        "       • Required: True                                                                                            \n",
-       "    • output_path (string): The path to save the extracted pages as a new PDF file.                                \n",
+       "    • end_page (integer): The last page to be extracted. This page is included in the extraction.                  \n",
        "       • Required: True                                                                                            \n",
+       "    • output_file (string): Name of the output file where the extracted pages will be saved.                       \n",
+       "       • Default: 'extracted_pages.pdf'                                                                            \n",
        " • Returns:                                                                                                        \n",
-       "    • output_path (string): The path to the extracted pages PDF file.                                              \n",
+       "    • output_file (string): Name of the output file where the extracted pages were saved.                          \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: extract_pdf_pages                                                                                         \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill extracts a specified section from a PDF file and saves it as a new PDF.                 \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: extract_pages_from_pdf                                                                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill extracts a specified range of pages from a PDF file and saves them as a new PDF file.   \n",
+       "\u001b[1;33m   \u001b[0mThe user needs to provide the path to the original PDF file and the range of pages to be extracted. The         \n",
+       "\u001b[1;33m   \u001b[0mextracted pages are saved in a new PDF file in the current working directory.                                   \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpdf_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m~/Downloads/voyager.pdf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[48;2;39;40;34m                                                                             \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mstart_page\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[48;2;39;40;34m                                                                                                   \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mend_page\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m5\u001b[0m\u001b[48;2;39;40;34m                                                                                                     \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34moutput_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34mextracted_pages.pdf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[48;2;39;40;34m                                                                              \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mextract_pdf_pages\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpdf_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mstart_page\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mend_page\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34moutput_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                   \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mextract_pages_from_pdf\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m~/Downloads/voyager.pdf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m5\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34mvoyager_extracted.pdf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mpdf_path\u001b[0m (string): The path to the PDF file.                                                                 \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mstart_page\u001b[0m (integer): The starting page number to extract.                                                   \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mpdf_path\u001b[0m (string): Path to the original PDF file.                                                            \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mend_page\u001b[0m (integer): The ending page number to extract.                                                       \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mstart_page\u001b[0m (integer): The first page to be extracted. Page numbers start from 1.                             \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1moutput_path\u001b[0m (string): The path to save the extracted pages as a new PDF file.                                \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mend_page\u001b[0m (integer): The last page to be extracted. This page is included in the extraction.                  \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1moutput_file\u001b[0m (string): Name of the output file where the extracted pages will be saved.                       \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mDefault: 'extracted_pages.pdf'                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1moutput_path\u001b[0m (string): The path to the extracted pages PDF file.                                              \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1moutput_file\u001b[0m (string): Name of the output file where the extracted pages were saved.                          \n"
       ]
      },
      "metadata": {},
@@ -571,13 +667,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 11,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "1025bc88f4c2439dbf1a24a060c7db54",
+       "model_id": "25767f6c233d4d3a986d5859754d2fed",
        "version_major": 2,
        "version_minor": 0
       },
@@ -607,6 +703,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/e71ac5c0-\n",
+       "685f-4f2b-9690-b77fb264eaec?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=90796;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/e71ac5c0-685f-4f2b-9690-b77fb264eaec?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/e71ac5c0-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=90796;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/e71ac5c0-685f-4f2b-9690-b77fb264eaec?poll=true\u001b\\\u001b[4;34m685f-4f2b-9690-b77fb264eaec?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -615,7 +728,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 12,
    "metadata": {},
    "outputs": [
     {
@@ -625,25 +738,13 @@
        "                                                  Skill Details:                                                   \n",
        "\n",
        " • Name: display_markdown_message                                                                                  \n",
-       " • Description: This skill is used to display a markdown message in a formatted way. It takes a multiline string as\n",
-       "   input and prints the message with proper formatting. It supports markdown syntax and can handle indentation and \n",
-       "   multiline strings.                                                                                              \n",
+       " • Description: This skill is used to display a markdown message. It works with multiline strings with lots of     \n",
+       "   indentation and will automatically make single line > tags beautiful.                                           \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " message = \"\"\"                                                                                                     \n",
-       " # Heading                                                                                                         \n",
-       "                                                                                                                   \n",
-       " - List item 1                                                                                                     \n",
-       " - List item 2                                                                                                     \n",
-       "                                                                                                                   \n",
-       " ---                                                                                                               \n",
-       "                                                                                                                   \n",
-       " > Blockquote                                                                                                      \n",
-       "                                                                                                                   \n",
-       " \"\"\"                                                                                                               \n",
-       " display_markdown_message(message)                                                                                 \n",
+       " display_markdown_message('> This is a markdown message.')                                                         \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
@@ -657,25 +758,13 @@
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: display_markdown_message                                                                                  \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is used to display a markdown message in a formatted way. It takes a multiline string as\n",
-       "\u001b[1;33m   \u001b[0minput and prints the message with proper formatting. It supports markdown syntax and can handle indentation and \n",
-       "\u001b[1;33m   \u001b[0mmultiline strings.                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is used to display a markdown message. It works with multiline strings with lots of     \n",
+       "\u001b[1;33m   \u001b[0mindentation and will automatically make single line > tags beautiful.                                           \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessage\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m\"\"\"\u001b[0m\u001b[48;2;39;40;34m                                                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m# Heading\u001b[0m\u001b[48;2;39;40;34m                                                                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m- List item 1\u001b[0m\u001b[48;2;39;40;34m                                                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m- List item 2\u001b[0m\u001b[48;2;39;40;34m                                                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m---\u001b[0m\u001b[48;2;39;40;34m                                                                                                              \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m> Blockquote\u001b[0m\u001b[48;2;39;40;34m                                                                                                     \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m\"\"\"\u001b[0m\u001b[48;2;39;40;34m                                                                                                              \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdisplay_markdown_message\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessage\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                                \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdisplay_markdown_message\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m> This is a markdown message.\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
@@ -702,13 +791,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 13,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "541ad5ff7e6146bbb08bde2c69149fe2",
+       "model_id": "4efa90f0f2004cd9beb5895cd879b702",
        "version_major": 2,
        "version_minor": 0
       },
@@ -738,6 +827,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/f3a428b6-\n",
+       "e8ef-4d61-b9a1-503afe355eaa?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=429238;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/f3a428b6-e8ef-4d61-b9a1-503afe355eaa?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/f3a428b6-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=429238;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/f3a428b6-e8ef-4d61-b9a1-503afe355eaa?poll=true\u001b\\\u001b[4;34me8ef-4d61-b9a1-503afe355eaa?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -746,7 +852,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 14,
    "metadata": {},
    "outputs": [
     {
@@ -755,20 +861,24 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: create_api                                                                                                \n",
-       " • Description: Generates a CodeSkill instance using different input sources.                                      \n",
+       " • Name: create                                                                                                    \n",
+       " • Description: This function generates a CodeSkill instance using different input sources. It can take in a       \n",
+       "   request string detailing the skill functionality, messages or a path to a JSON file containing messages, a      \n",
+       "   string of file content or path to a code/API doc file, a directory path with skill name as stem or file path    \n",
+       "   with skill.json as stem, an identifier for a Huggingface repository, or a path to the skill within the          \n",
+       "   Huggingface repository.                                                                                         \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " create_api(request, messages=messages, messages_json_path=messages_json_path, file_content=file_content,          \n",
-       " file_path=file_path, skill_path=skill_path, skill_json_path=skill_json_path,                                      \n",
-       " huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path)                           \n",
+       " from creator import create                                                                                        \n",
+       "                                                                                                                   \n",
+       " skill = create(request='...', messages=[...], file_content='...', file_path='...', skill_path='...',              \n",
+       " skill_json_path='...', huggingface_repo_id='...', huggingface_skill_path='...')                                   \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
        "    • request (string): String detailing the skill functionality.                                                  \n",
-       "       • Required: True                                                                                            \n",
        "    • messages (array): Messages as a list of dictionaries.                                                        \n",
        "    • messages_json_path (string): Path to a JSON file containing messages.                                        \n",
        "    • file_content (string): String of file content.                                                               \n",
@@ -785,20 +895,24 @@
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: create_api                                                                                                \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: Generates a \u001b[1;36;40mCodeSkill\u001b[0m instance using different input sources.                                      \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: create                                                                                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This function generates a \u001b[1;36;40mCodeSkill\u001b[0m instance using different input sources. It can take in a       \n",
+       "\u001b[1;33m   \u001b[0mrequest string detailing the skill functionality, messages or a path to a JSON file containing messages, a      \n",
+       "\u001b[1;33m   \u001b[0mstring of file content or path to a code/API doc file, a directory path with skill name as stem or file path    \n",
+       "\u001b[1;33m   \u001b[0mwith \u001b[1;36;40mskill.json\u001b[0m as stem, an identifier for a Huggingface repository, or a path to the skill within the          \n",
+       "\u001b[1;33m   \u001b[0mHuggingface repository.                                                                                         \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate_api\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mrequest\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages_json_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                          \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34mfrom\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreator\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34mimport\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate\u001b[0m\u001b[48;2;39;40;34m                                                                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mrequest\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m            \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                  \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mrequest\u001b[0m (string): String detailing the skill functionality.                                                  \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mmessages\u001b[0m (array): Messages as a list of dictionaries.                                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mmessages_json_path\u001b[0m (string): Path to a JSON file containing messages.                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mfile_content\u001b[0m (string): String of file content.                                                               \n",
@@ -833,19 +947,19 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": 15,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "
\n",
-       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api                                        \n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create                                            \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api\u001b[0m\u001b[35m                                        \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create\u001b[0m\u001b[35m                                            \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -862,7 +976,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 19,
+   "execution_count": 16,
    "metadata": {},
    "outputs": [
     {

From 9f5f696dae34d82277a9eef19253efafdd8b943f Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 15:30:01 +0800
Subject: [PATCH 43/56] fix: runnable bugs

---
 creator/core/runnable.py                    |   7 +-
 creator/core/skill.py                       |   5 +-
 creator/retrivever/base.py                  |   2 +-
 docs/examples/02_skills_library.ipynb       |  82 ++++---
 docs/examples/03_skills_search.ipynb        |  11 +-
 docs/examples/04_skills_run.ipynb           | 230 ++++++++++++-------
 docs/examples/05_skills_test.ipynb          | 122 ++++++-----
 docs/examples/06_skills_refactor.ipynb      | 231 +++++++++++++++-----
 docs/examples/07_skills_auto_optimize.ipynb | 150 +++++++++++--
 docs/examples/08_creator_agent.ipynb        | 169 +++++++++-----
 10 files changed, 709 insertions(+), 300 deletions(-)

diff --git a/creator/core/runnable.py b/creator/core/runnable.py
index da64ad6..2b7da10 100644
--- a/creator/core/runnable.py
+++ b/creator/core/runnable.py
@@ -121,7 +121,6 @@ def construct_test_skill_messages(inputs):
     return {"messages": messages}
 
 
-@print_run_url
 @runnable(run_name="TestSkill")
 def test_skill(inputs, config: RunnableConfig):
     code_tester_agent = create_code_tester_agent(creator_config)
@@ -156,7 +155,6 @@ def construct_refactor_skill_messages(inputs):
     return {"messages": messages}
 
 
-@print_run_url
 @runnable(run_name="RefactorSkill")
 def refactor_skill(inputs, config: RunnableConfig):
     code_refactor_agent = create_code_refactor_agent(creator_config)
@@ -182,6 +180,10 @@ def auto_optimize_skill(inputs, config: RunnableConfig):
             conversation_history = conversation_history + test_result["messages"]
             if "test_summary" in test_result:
                 test_summary = test_result["test_summary"]
+                if isinstance(test_summary, dict) and "test_cases" in test_summary:
+                    pass
+                else:
+                    test_summary = {"test_cases": test_summary}
                 all_passed = all(test_case["is_passed"] for test_case in test_summary["test_cases"])
                 if all_passed and refined:
                     skill.conversation_history = conversation_history
@@ -190,7 +192,6 @@ def auto_optimize_skill(inputs, config: RunnableConfig):
                         "test_summary": test_summary,
                     }
             print(f"> Auto Refine Skill {i+1}/{retry_times}", print_type="markdown")
-            skill.Config.runnable_config = config
             skill = skill > "I have tested the skill, but it failed, please refine it."
             skill.conversation_history = conversation_history
             if all_passed:
diff --git a/creator/core/skill.py b/creator/core/skill.py
index c9009f7..1c4573f 100644
--- a/creator/core/skill.py
+++ b/creator/core/skill.py
@@ -148,7 +148,6 @@ class Config:
         skills_to_combine = []
         user_request = "please help me refine the skill object"
         refactor_type = "Refine"
-        runnable_config = None
 
     def __init__(self, **data):
         super().__init__(**data)
@@ -282,7 +281,7 @@ def refactor(self):
             "skill_program_language": self.skill_program_language,
             "skill_code": self.skill_code,
             "user_request": self.Config.user_request,
-        }, self.Config.runnable_config)
+        })
         refactored_skills = []
         for refactored_skill_json in refactored_skill_jsons:
             refactored_skill = CodeSkill(**refactored_skill_json)
@@ -299,7 +298,7 @@ def auto_optimize(self, retry_times=3):
         optimized_result = auto_optimize_skill.invoke({"old_skill": self, "retry_times": retry_times})
         test_summary = optimized_result["test_summary"]
         skill = optimized_result["skill"]
-        if isinstance(test_summary, None):
+        if test_summary is None:
             return skill
         elif isinstance(test_summary, TestSummary):
             skill.test_summary = test_summary
diff --git a/creator/retrivever/base.py b/creator/retrivever/base.py
index 42cfc62..1f4a4b7 100644
--- a/creator/retrivever/base.py
+++ b/creator/retrivever/base.py
@@ -32,7 +32,7 @@ def index(self, documents: List[Any], ids: List[str] = None, metadatas: List[Dic
             metadatas = documents
         texts = [self._preprocess(doc) for doc in documents]
         if ids is None:
-            ids = [generate_uuid_like_string() for _ in range(len(texts))]
+            ids = [generate_uuid_like_string(text) for text in texts]
         if self.db is None:
             self.db = Qdrant.from_texts(texts=texts, embedding=self.embedding, metadatas=metadatas, ids=ids, path=self.vectordb_path, collection_name=self.collection_name)
         else:
diff --git a/docs/examples/02_skills_library.ipynb b/docs/examples/02_skills_library.ipynb
index f88ff36..7d3aec6 100644
--- a/docs/examples/02_skills_library.ipynb
+++ b/docs/examples/02_skills_library.ipynb
@@ -59,7 +59,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "d19fad8112b548a18fc1a1a0e05a246d",
+       "model_id": "0fad0f77899147108a9a42beeab4bfc9",
        "version_major": 2,
        "version_minor": 0
       },
@@ -89,6 +89,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b898dfdb-\n",
+       "c596-4538-b4c9-d88136c3fb4f?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=81787;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b898dfdb-c596-4538-b4c9-d88136c3fb4f?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b898dfdb-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=81787;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b898dfdb-c596-4538-b4c9-d88136c3fb4f?poll=true\u001b\\\u001b[4;34mc596-4538-b4c9-d88136c3fb4f?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -106,20 +123,24 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: create_api                                                                                                \n",
-       " • Description: Generates a CodeSkill instance using different input sources.                                      \n",
+       " • Name: create                                                                                                    \n",
+       " • Description: This function generates a CodeSkill instance using different input sources. It can take in a       \n",
+       "   request string detailing the skill functionality, messages or a path to a JSON file containing messages, a      \n",
+       "   string of file content or path to a code/API doc file, a directory path with skill name as stem or file path    \n",
+       "   with skill.json as stem, an identifier for a Huggingface repository, or a path to the skill within the          \n",
+       "   Huggingface repository.                                                                                         \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " create_api(request, messages=messages, messages_json_path=messages_json_path, file_content=file_content,          \n",
-       " file_path=file_path, skill_path=skill_path, skill_json_path=skill_json_path,                                      \n",
-       " huggingface_repo_id=huggingface_repo_id, huggingface_skill_path=huggingface_skill_path)                           \n",
+       " from creator import create                                                                                        \n",
+       "                                                                                                                   \n",
+       " skill = create(request='...', messages=[...], file_content='...', file_path='...', skill_path='...',              \n",
+       " skill_json_path='...', huggingface_repo_id='...', huggingface_skill_path='...')                                   \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
        "    • request (string): String detailing the skill functionality.                                                  \n",
-       "       • Required: True                                                                                            \n",
        "    • messages (array): Messages as a list of dictionaries.                                                        \n",
        "    • messages_json_path (string): Path to a JSON file containing messages.                                        \n",
        "    • file_content (string): String of file content.                                                               \n",
@@ -136,20 +157,24 @@
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: create_api                                                                                                \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: Generates a \u001b[1;36;40mCodeSkill\u001b[0m instance using different input sources.                                      \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: create                                                                                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This function generates a \u001b[1;36;40mCodeSkill\u001b[0m instance using different input sources. It can take in a       \n",
+       "\u001b[1;33m   \u001b[0mrequest string detailing the skill functionality, messages or a path to a JSON file containing messages, a      \n",
+       "\u001b[1;33m   \u001b[0mstring of file content or path to a code/API doc file, a directory path with skill name as stem or file path    \n",
+       "\u001b[1;33m   \u001b[0mwith \u001b[1;36;40mskill.json\u001b[0m as stem, an identifier for a Huggingface repository, or a path to the skill within the          \n",
+       "\u001b[1;33m   \u001b[0mHuggingface repository.                                                                                         \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate_api\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mrequest\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages_json_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                          \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34mfrom\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreator\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34mimport\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate\u001b[0m\u001b[48;2;39;40;34m                                                                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcreate\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mrequest\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mmessages\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m.\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_content\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mfile_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m            \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mskill_json_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_repo_id\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mhuggingface_skill_path\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m...\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                  \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mrequest\u001b[0m (string): String detailing the skill functionality.                                                  \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mmessages\u001b[0m (array): Messages as a list of dictionaries.                                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mmessages_json_path\u001b[0m (string): Path to a JSON file containing messages.                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mfile_content\u001b[0m (string): String of file content.                                                               \n",
@@ -179,12 +204,12 @@
      "data": {
       "text/html": [
        "
\n",
-       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api                                        \n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create                                            \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api\u001b[0m\u001b[35m                                        \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create\u001b[0m\u001b[35m                                            \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -204,19 +229,19 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "/Users/gongjunmin/.cache/open_creator/remote/ChuxiJ/skill_library/create_api\n"
+      "/Users/gongjunmin/.cache/open_creator/remote/ChuxiJ/skill_library/create\n"
      ]
     },
     {
      "data": {
       "text/html": [
        "
\n",
-       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api                                        \n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create                                            \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api\u001b[0m\u001b[35m                                        \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create\u001b[0m\u001b[35m                                            \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -236,11 +261,12 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n"
+      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_quadratic_equation\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m               \u001b[1m\u001b[36msolve_random_maze\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_scatter_plot\u001b[m\u001b[m      \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n"
      ]
     }
    ],
@@ -265,12 +291,12 @@
      "data": {
       "text/html": [
        "
\n",
-       "▌ saved to /Users/gongjunmin/.cache/open_creator/remote/ChuxiJ/skill_library/create_api/create_api               \n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api                                        \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/remote/ChuxiJ/skill_library/create_api/create_api\u001b[0m\u001b[35m               \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/create_api\u001b[0m\u001b[35m                                        \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -388,14 +414,12 @@
      "data": {
       "text/html": [
        "
\n",
-       "▌ saved to                                                                                                       \n",
-       "▌ /Users/gongjunmin/.cache/open_creator/remote/Sayoyo/skill-library/extract_pdf_section/extract_pdf_section      \n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/extract_pdf_section                               \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to \u001b[0m\u001b[35m                                                                                                      \u001b[0m\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35m/Users/gongjunmin/.cache/open_creator/remote/Sayoyo/skill-library/extract_pdf_section/extract_pdf_section\u001b[0m\u001b[35m      \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/extract_pdf_section\u001b[0m\u001b[35m                               \u001b[0m\n"
       ]
      },
      "metadata": {},
diff --git a/docs/examples/03_skills_search.ipynb b/docs/examples/03_skills_search.ipynb
index 72b2a7b..de5ffca 100644
--- a/docs/examples/03_skills_search.ipynb
+++ b/docs/examples/03_skills_search.ipynb
@@ -52,11 +52,12 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n"
+      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_quadratic_equation\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m               \u001b[1m\u001b[36msolve_random_maze\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_scatter_plot\u001b[m\u001b[m      \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n"
      ]
     }
    ],
diff --git a/docs/examples/04_skills_run.ipynb b/docs/examples/04_skills_run.ipynb
index a4d26d0..0a6aff5 100644
--- a/docs/examples/04_skills_run.ipynb
+++ b/docs/examples/04_skills_run.ipynb
@@ -38,7 +38,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "../docs/tech_report/open-creator.pdf\n"
+      "ls: ./tech_report/open-creator.pdf: No such file or directory\n"
      ]
     }
    ],
@@ -48,9 +48,24 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 3,
    "metadata": {},
    "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "
\n",
+       "▌ loading vector database...                                                                                     \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mloading vector database...\u001b[0m\u001b[35m                                                                                     \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -118,7 +133,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 4,
    "metadata": {},
    "outputs": [
     {
@@ -184,7 +199,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -211,7 +226,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -225,7 +240,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [
     {
@@ -234,7 +249,7 @@
        "'./data/open-creator2-5.pdf'"
       ]
      },
-     "execution_count": 10,
+     "execution_count": 7,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -252,19 +267,19 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "
\n",
-       "▌ loading vector database...                                                                                     \n",
+       "▌ Installing dependencies                                                                                        \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35mloading vector database...\u001b[0m\u001b[35m                                                                                     \u001b[0m\n"
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstalling dependencies\u001b[0m\u001b[35m                                                                                        \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -273,19 +288,15 @@
     {
      "data": {
       "text/html": [
-       "
{\n",
-       "  \"status\": \"success\",\n",
-       "  \"stdout\": \"\",\n",
-       "  \"stderr\": \"\"\n",
-       "}\n",
+       "
                                                                                                                   \n",
+       " pip show PyPDF2 || pip install \"PyPDF2>=1.26.0\"                                                                   \n",
+       "                                                                                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[1m{\u001b[0m\n",
-       "  \u001b[1;34m\"status\"\u001b[0m: \u001b[32m\"success\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stdout\"\u001b[0m: \u001b[32m\"\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stderr\"\u001b[0m: \u001b[32m\"\"\u001b[0m\n",
-       "\u001b[1m}\u001b[0m\n"
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mshow\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mPyPDF2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m||\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minstall\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m\"PyPDF2>=1.26.0\"\u001b[0m\u001b[48;2;39;40;34m                                                                  \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -293,13 +304,22 @@
     },
     {
      "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "decbe223ddf0462b99a8819c2891d198",
-       "version_major": 2,
-       "version_minor": 0
-      },
+      "text/html": [
+       "
\n",
+       "▌ Install dependencies result: {'status': 'success', 'stdout': 'Name: PyPDF2\\nVersion: 3.0.1\\nSummary: A         \n",
+       "▌ pure-python PDF library capable of splitting, merging, cropping, and transforming PDF files\\nHome-page:        \n",
+       "▌ \\nAuthor: \\nAuthor-email: Mathieu Fenniak biziqe@mathieu.fenniak.net\\nLicense: \\nLocation:                     \n",
+       "▌ /Users/gongjunmin/miniconda3/envs/open_creator_online/lib/python3.10/site-packages\\nRequires: \\nRequired-by:   \n",
+       "▌ \\n', 'stderr': ''}                                                                                             \n",
+       "
\n"
+      ],
       "text/plain": [
-       "Output()"
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstall dependencies result: {'status': 'success', 'stdout': 'Name: PyPDF2\\nVersion: 3.0.1\\nSummary: A \u001b[0m\u001b[35m        \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mpure-python PDF library capable of splitting, merging, cropping, and transforming PDF files\\nHome-page: \u001b[0m\u001b[35m       \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m\\nAuthor: \\nAuthor-email: Mathieu Fenniak \u001b[0m\u001b]8;id=463524;mailto:biziqe@mathieu.fenniak.net\u001b\\\u001b[4;34mbiziqe@mathieu.fenniak.net\u001b[0m\u001b]8;;\u001b\\\u001b[35m\\nLicense: \\nLocation: \u001b[0m\u001b[35m                    \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m/Users/gongjunmin/miniconda3/envs/open_creator_online/lib/python3.10/site-packages\\nRequires: \\nRequired-by: \u001b[0m\u001b[35m  \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m\\n', 'stderr': ''}\u001b[0m\u001b[35m                                                                                             \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -308,7 +328,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "6b9f6a7914a541268042529481a0af39",
+       "model_id": "3b2d13db34d543c29e28eca4539ff644",
        "version_major": 2,
        "version_minor": 0
       },
@@ -342,7 +362,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "1b5d53466fbf4132938cb62f06f948ee",
+       "model_id": "72d9c1b7db8449c899744345865925f4",
        "version_major": 2,
        "version_minor": 0
       },
@@ -373,6 +393,23 @@
      "metadata": {},
      "output_type": "display_data"
     },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/8e9f879c-\n",
+       "1112-4a54-aa62-e1430ccddad6?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=340957;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/8e9f879c-1112-4a54-aa62-e1430ccddad6?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/8e9f879c-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=340957;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/8e9f879c-1112-4a54-aa62-e1430ccddad6?poll=true\u001b\\\u001b[4;34m1112-4a54-aa62-e1430ccddad6?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -397,30 +434,29 @@
        "  },\n",
        "  {\n",
        "    \"role\": \"user\",\n",
-       "    \"content\": \"{\\\"pdf_path\\\": \\\"../docs/tech_report/open-creator.pdf\\\", \\\"start_page\\\": 2, \\\"end_page\\\": 5, \n",
+       "    \"content\": \"{\\\"pdf_path\\\": \\\"../tech_report/open-creator.pdf\\\", \\\"start_page\\\": 2, \\\"end_page\\\": 5, \n",
        "\\\"output_path\\\": \\\"./data/open-creator2-5.pdf\\\"}\"\n",
        "  },\n",
        "  {\n",
        "    \"role\": \"assistant\",\n",
-       "    \"content\": \"To extract pages 2 to 5 from the PDF file \\\"../docs/tech_report/open-creator.pdf\\\" and save them as\n",
-       "a new file at \\\"./data/open-creator2-5.pdf\\\", I will use the `extract_pdf_section` function.\",\n",
+       "    \"content\": null,\n",
        "    \"function_call\": {\n",
        "      \"name\": \"run_code\",\n",
        "      \"arguments\": \"{\\\"language\\\": \\\"python\\\", \\\"code\\\": \\\"{\\\\\\\"language\\\\\\\": \\\\\\\"python\\\\\\\", \\\\\\\"code\\\\\\\": \n",
-       "\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\n",
+       "\\\\\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\n",
        "    }\n",
        "  },\n",
        "  {\n",
        "    \"role\": \"function\",\n",
        "    \"content\": \"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"{'language': 'python', 'code': \n",
-       "\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\\n\\\", \n",
-       "\\\"stderr\\\": \\\"\\\"}\",\n",
+       "\\\\\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\", \\\"stderr\\\": \n",
+       "\\\"\\\"}\",\n",
        "    \"name\": \"run_code\"\n",
        "  },\n",
        "  {\n",
        "    \"role\": \"assistant\",\n",
-       "    \"content\": \"The pages 2 to 5 have been extracted from the PDF file and saved as a new file at \n",
-       "\\\"./data/open-creator2-5.pdf\\\".\"\n",
+       "    \"content\": \"The code has been executed successfully. The PDF section has been extracted and saved to the \n",
+       "specified output path. You can now access the extracted section from the output file.\"\n",
        "  }\n",
        "]\n",
        "
\n"
@@ -447,30 +483,29 @@
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"user\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"{\\\"pdf_path\\\": \\\"../docs/tech_report/open-creator.pdf\\\", \\\"start_page\\\": 2, \\\"end_page\\\": 5, \u001b[0m\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"{\\\"pdf_path\\\": \\\"../tech_report/open-creator.pdf\\\", \\\"start_page\\\": 2, \\\"end_page\\\": 5, \u001b[0m\n",
        "\u001b[32m\\\"output_path\\\": \\\"./data/open-creator2-5.pdf\\\"}\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"assistant\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"To extract pages 2 to 5 from the PDF file \\\"../docs/tech_report/open-creator.pdf\\\" and save them as\u001b[0m\n",
-       "\u001b[32ma new file at \\\"./data/open-creator2-5.pdf\\\", I will use the `extract_pdf_section` function.\"\u001b[0m,\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[3;35mnull\u001b[0m,\n",
        "    \u001b[1;34m\"function_call\"\u001b[0m: \u001b[1m{\u001b[0m\n",
        "      \u001b[1;34m\"name\"\u001b[0m: \u001b[32m\"run_code\"\u001b[0m,\n",
        "      \u001b[1;34m\"arguments\"\u001b[0m: \u001b[32m\"{\\\"language\\\": \\\"python\\\", \\\"code\\\": \\\"{\\\\\\\"language\\\\\\\": \\\\\\\"python\\\\\\\", \\\\\\\"code\\\\\\\": \u001b[0m\n",
-       "\u001b[32m\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\u001b[0m\n",
+       "\u001b[32m\\\\\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\u001b[0m\n",
        "    \u001b[1m}\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"function\"\u001b[0m,\n",
        "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"{'language': 'python', 'code': \u001b[0m\n",
-       "\u001b[32m\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\\n\\\", \u001b[0m\n",
-       "\u001b[32m\\\"stderr\\\": \\\"\\\"}\"\u001b[0m,\n",
+       "\u001b[32m\\\\\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\", \\\"stderr\\\": \u001b[0m\n",
+       "\u001b[32m\\\"\\\"}\"\u001b[0m,\n",
        "    \u001b[1;34m\"name\"\u001b[0m: \u001b[32m\"run_code\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"assistant\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"The pages 2 to 5 have been extracted from the PDF file and saved as a new file at \u001b[0m\n",
-       "\u001b[32m\\\"./data/open-creator2-5.pdf\\\".\"\u001b[0m\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"The code has been executed successfully. The PDF section has been extracted and saved to the \u001b[0m\n",
+       "\u001b[32mspecified output path. You can now access the extracted section from the output file.\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m\n",
        "\u001b[1m]\u001b[0m\n"
       ]
@@ -503,25 +538,59 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 9,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
-       "
{\n",
-       "  \"status\": \"success\",\n",
-       "  \"stdout\": \"\",\n",
-       "  \"stderr\": \"\"\n",
-       "}\n",
+       "
\n",
+       "▌ Installing dependencies                                                                                        \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstalling dependencies\u001b[0m\u001b[35m                                                                                        \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                                                                                                                   \n",
+       " pip show PyPDF2 || pip install \"PyPDF2>=1.26.0\"                                                                   \n",
+       "                                                                                                                   \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mshow\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mPyPDF2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m||\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minstall\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m\"PyPDF2>=1.26.0\"\u001b[0m\u001b[48;2;39;40;34m                                                                  \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
\n",
+       "▌ Install dependencies result: {'status': 'success', 'stdout': 'Name: PyPDF2\\nVersion: 3.0.1\\nSummary: A         \n",
+       "▌ pure-python PDF library capable of splitting, merging, cropping, and transforming PDF files\\nHome-page:        \n",
+       "▌ \\nAuthor: \\nAuthor-email: Mathieu Fenniak biziqe@mathieu.fenniak.net\\nLicense: \\nLocation:                     \n",
+       "▌ /Users/gongjunmin/miniconda3/envs/open_creator_online/lib/python3.10/site-packages\\nRequires: \\nRequired-by:   \n",
+       "▌ \\n', 'stderr': ''}                                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[1m{\u001b[0m\n",
-       "  \u001b[1;34m\"status\"\u001b[0m: \u001b[32m\"success\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stdout\"\u001b[0m: \u001b[32m\"\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stderr\"\u001b[0m: \u001b[32m\"\"\u001b[0m\n",
-       "\u001b[1m}\u001b[0m\n"
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstall dependencies result: {'status': 'success', 'stdout': 'Name: PyPDF2\\nVersion: 3.0.1\\nSummary: A \u001b[0m\u001b[35m        \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mpure-python PDF library capable of splitting, merging, cropping, and transforming PDF files\\nHome-page: \u001b[0m\u001b[35m       \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m\\nAuthor: \\nAuthor-email: Mathieu Fenniak \u001b[0m\u001b]8;id=265924;mailto:biziqe@mathieu.fenniak.net\u001b\\\u001b[4;34mbiziqe@mathieu.fenniak.net\u001b[0m\u001b]8;;\u001b\\\u001b[35m\\nLicense: \\nLocation: \u001b[0m\u001b[35m                    \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m/Users/gongjunmin/miniconda3/envs/open_creator_online/lib/python3.10/site-packages\\nRequires: \\nRequired-by: \u001b[0m\u001b[35m  \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35m\\n', 'stderr': ''}\u001b[0m\u001b[35m                                                                                             \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -530,7 +599,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "fc4e1f4dd91642c5aef19c8022ffb169",
+       "model_id": "42950aa643bc4ef68fa89b28c76938b0",
        "version_major": 2,
        "version_minor": 0
       },
@@ -564,7 +633,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "8f0cd6de709744018cf3f00241094e0f",
+       "model_id": "89e7e660999d4305bc57ae5575a1c250",
        "version_major": 2,
        "version_minor": 0
       },
@@ -595,6 +664,23 @@
      "metadata": {},
      "output_type": "display_data"
     },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6d48421f-\n",
+       "b99e-41f3-b794-14d4a553fd1b?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=124701;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6d48421f-b99e-41f3-b794-14d4a553fd1b?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6d48421f-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=124701;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6d48421f-b99e-41f3-b794-14d4a553fd1b?poll=true\u001b\\\u001b[4;34mb99e-41f3-b794-14d4a553fd1b?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -619,7 +705,7 @@
        "  },\n",
        "  {\n",
        "    \"role\": \"user\",\n",
-       "    \"content\": \"extract 2-5 pages section from pdf path '../docs/tech_report/open-creator.pdf' to \n",
+       "    \"content\": \"extract 2-5 pages section from pdf path '../tech_report/open-creator.pdf' to \n",
        "'./data/open-creator2-5.pdf'\"\n",
        "  },\n",
        "  {\n",
@@ -627,21 +713,19 @@
        "    \"content\": null,\n",
        "    \"function_call\": {\n",
        "      \"name\": \"run_code\",\n",
-       "      \"arguments\": \"{\\\"language\\\": \\\"python\\\", \\\"code\\\": \\\"{\\\\\\\"language\\\\\\\": \\\\\\\"python\\\\\\\", \\\\\\\"code\\\\\\\": \n",
-       "\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\n",
+       "      \"arguments\": \"{\\n\\\"language\\\": \\\"python\\\", \\n\\\"code\\\": \n",
+       "\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\"\\n}\"\n",
        "    }\n",
        "  },\n",
        "  {\n",
        "    \"role\": \"function\",\n",
-       "    \"content\": \"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"{'language': 'python', 'code': \n",
-       "\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\\n\\\", \n",
-       "\\\"stderr\\\": \\\"\\\"}\",\n",
+       "    \"content\": \"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"./data/open-creator2-5.pdf\\\", \\\"stderr\\\": \\\"\\\"}\",\n",
        "    \"name\": \"run_code\"\n",
        "  },\n",
        "  {\n",
        "    \"role\": \"assistant\",\n",
-       "    \"content\": \"The PDF section has been extracted successfully. You can download the extracted section from \n",
-       "[here](sandbox:/Users/gongjunmin/LLM/open_creator_dev/open-creator/examples/data/open-creator2-5.pdf).\"\n",
+       "    \"content\": \"The extraction of pages 2-5 from the PDF at '../tech_report/open-creator.pdf' has been successful. \n",
+       "The extracted section has been saved to './data/open-creator2-5.pdf'.\"\n",
        "  }\n",
        "]\n",
        "
\n"
@@ -668,7 +752,7 @@
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"user\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"extract 2-5 pages section from pdf path '../docs/tech_report/open-creator.pdf' to \u001b[0m\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"extract 2-5 pages section from pdf path '../tech_report/open-creator.pdf' to \u001b[0m\n",
        "\u001b[32m'./data/open-creator2-5.pdf'\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
@@ -676,21 +760,19 @@
        "    \u001b[1;34m\"content\"\u001b[0m: \u001b[3;35mnull\u001b[0m,\n",
        "    \u001b[1;34m\"function_call\"\u001b[0m: \u001b[1m{\u001b[0m\n",
        "      \u001b[1;34m\"name\"\u001b[0m: \u001b[32m\"run_code\"\u001b[0m,\n",
-       "      \u001b[1;34m\"arguments\"\u001b[0m: \u001b[32m\"{\\\"language\\\": \\\"python\\\", \\\"code\\\": \\\"{\\\\\\\"language\\\\\\\": \\\\\\\"python\\\\\\\", \\\\\\\"code\\\\\\\": \u001b[0m\n",
-       "\u001b[32m\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\"}\"\u001b[0m\n",
+       "      \u001b[1;34m\"arguments\"\u001b[0m: \u001b[32m\"{\\n\\\"language\\\": \\\"python\\\", \\n\\\"code\\\": \u001b[0m\n",
+       "\u001b[32m\\\"extract_pdf_section('../tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\"\\n}\"\u001b[0m\n",
        "    \u001b[1m}\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"function\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"{'language': 'python', 'code': \u001b[0m\n",
-       "\u001b[32m\\\\\\\"extract_pdf_section('../docs/tech_report/open-creator.pdf', 2, 5, './data/open-creator2-5.pdf')\\\\\\\"}\\\\n\\\", \u001b[0m\n",
-       "\u001b[32m\\\"stderr\\\": \\\"\\\"}\"\u001b[0m,\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"{\\\"status\\\": \\\"success\\\", \\\"stdout\\\": \\\"./data/open-creator2-5.pdf\\\", \\\"stderr\\\": \\\"\\\"}\"\u001b[0m,\n",
        "    \u001b[1;34m\"name\"\u001b[0m: \u001b[32m\"run_code\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m,\n",
        "  \u001b[1m{\u001b[0m\n",
        "    \u001b[1;34m\"role\"\u001b[0m: \u001b[32m\"assistant\"\u001b[0m,\n",
-       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"The PDF section has been extracted successfully. You can download the extracted section from \u001b[0m\n",
-       "\u001b[32m[here](sandbox:/Users/gongjunmin/LLM/open_creator_dev/open-creator/examples/data/open-creator2-5.pdf).\"\u001b[0m\n",
+       "    \u001b[1;34m\"content\"\u001b[0m: \u001b[32m\"The extraction of pages 2-5 from the PDF at '../tech_report/open-creator.pdf' has been successful. \u001b[0m\n",
+       "\u001b[32mThe extracted section has been saved to './data/open-creator2-5.pdf'.\"\u001b[0m\n",
        "  \u001b[1m}\u001b[0m\n",
        "\u001b[1m]\u001b[0m\n"
       ]
@@ -703,7 +785,7 @@
     "skills = search(\"pdf extract section\")\n",
     "if skills:\n",
     "  skill = skills[0]\n",
-    "  request = \"extract 2-5 pages section from pdf path '../docs/tech_report/open-creator.pdf' to './data/open-creator2-5.pdf'\"\n",
+    "  request = \"extract 2-5 pages section from pdf path '../tech_report/open-creator.pdf' to './data/open-creator2-5.pdf'\"\n",
     "  messages = skill.run(request)\n",
     "  print(messages, print_type=\"json\")"
    ]
diff --git a/docs/examples/05_skills_test.ipynb b/docs/examples/05_skills_test.ipynb
index 8429294..aae434c 100644
--- a/docs/examples/05_skills_test.ipynb
+++ b/docs/examples/05_skills_test.ipynb
@@ -29,11 +29,12 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m\n",
-      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n"
+      "\u001b[1m\u001b[36mask_run_code_confirm\u001b[m\u001b[m     \u001b[1m\u001b[36mextract_pdf_section\u001b[m\u001b[m      \u001b[1m\u001b[36msolve_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcount_prime_numbers\u001b[m\u001b[m      \u001b[1m\u001b[36mextract_section_from_pdf\u001b[m\u001b[m \u001b[1m\u001b[36msolve_game_of_24\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate\u001b[m\u001b[m                   \u001b[1m\u001b[36mfilter_prime_numbers\u001b[m\u001b[m     \u001b[1m\u001b[36msolve_quadratic_equation\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_api\u001b[m\u001b[m               \u001b[1m\u001b[36mgame_of_24\u001b[m\u001b[m               \u001b[1m\u001b[36msolve_random_maze\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mcreate_scatter_plot\u001b[m\u001b[m      \u001b[1m\u001b[36mgame_of_24_solver\u001b[m\u001b[m\n",
+      "\u001b[1m\u001b[36mdisplay_markdown_message\u001b[m\u001b[m \u001b[1m\u001b[36mlist_python_functions\u001b[m\u001b[m\n"
      ]
     }
    ],
@@ -113,13 +114,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "b6ada793fc744a30934de8526909bab7",
+       "model_id": "747b76bb1aba4f8c930caeb9a40797de",
        "version_major": 2,
        "version_minor": 0
       },
@@ -133,7 +134,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "b612d423a0b149aa950a0a0d0dd70aa6",
+       "model_id": "3c838524ec9042ce85d51eca088af2f4",
        "version_major": 2,
        "version_minor": 0
       },
@@ -167,7 +168,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "25fcc1f76d314cce9a7ad35776b64989",
+       "model_id": "edfeec012c7d42d9b2d69cf9252f2fba",
        "version_major": 2,
        "version_minor": 0
       },
@@ -197,6 +198,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/2bff98db-\n",
+       "7d63-4608-a487-ea7ec12f05a2?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=883877;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/2bff98db-7d63-4608-a487-ea7ec12f05a2?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/2bff98db-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=883877;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/2bff98db-7d63-4608-a487-ea7ec12f05a2?poll=true\u001b\\\u001b[4;34m7d63-4608-a487-ea7ec12f05a2?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -206,7 +224,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [
     {
@@ -217,46 +235,46 @@
        "\n",
        "                                                    Test Case 0                                                    \n",
        "\n",
-       " • Test Input: (2, 2)                                                                                              \n",
-       " • Run Command: filter_prime_numbers(2, 2)                                                                         \n",
-       " • Expected Result: 1                                                                                              \n",
-       " • Actual Result: 1                                                                                                \n",
+       " • Test Input: (2, 10)                                                                                             \n",
+       " • Run Command: filter_prime_numbers(2, 10)                                                                        \n",
+       " • Expected Result: 4                                                                                              \n",
+       " • Actual Result: 4                                                                                                \n",
        " • Is Passed: Yes                                                                                                  \n",
        "\n",
        "───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
        "                                                    Test Case 1                                                    \n",
        "\n",
-       " • Test Input: (2, 3)                                                                                              \n",
-       " • Run Command: filter_prime_numbers(2, 3)                                                                         \n",
-       " • Expected Result: 2                                                                                              \n",
-       " • Actual Result: 2                                                                                                \n",
+       " • Test Input: (11, 20)                                                                                            \n",
+       " • Run Command: filter_prime_numbers(11, 20)                                                                       \n",
+       " • Expected Result: 4                                                                                              \n",
+       " • Actual Result: 4                                                                                                \n",
        " • Is Passed: Yes                                                                                                  \n",
        "\n",
        "───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
        "                                                    Test Case 2                                                    \n",
        "\n",
-       " • Test Input: (2, 10)                                                                                             \n",
-       " • Run Command: filter_prime_numbers(2, 10)                                                                        \n",
-       " • Expected Result: 4                                                                                              \n",
-       " • Actual Result: 4                                                                                                \n",
+       " • Test Input: (21, 30)                                                                                            \n",
+       " • Run Command: filter_prime_numbers(21, 30)                                                                       \n",
+       " • Expected Result: 2                                                                                              \n",
+       " • Actual Result: 2                                                                                                \n",
        " • Is Passed: Yes                                                                                                  \n",
        "\n",
        "───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
        "                                                    Test Case 3                                                    \n",
        "\n",
-       " • Test Input: (2, 20)                                                                                             \n",
-       " • Run Command: filter_prime_numbers(2, 20)                                                                        \n",
-       " • Expected Result: 8                                                                                              \n",
-       " • Actual Result: 8                                                                                                \n",
+       " • Test Input: (31, 40)                                                                                            \n",
+       " • Run Command: filter_prime_numbers(31, 40)                                                                       \n",
+       " • Expected Result: 2                                                                                              \n",
+       " • Actual Result: 2                                                                                                \n",
        " • Is Passed: Yes                                                                                                  \n",
        "\n",
        "───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
        "                                                    Test Case 4                                                    \n",
        "\n",
-       " • Test Input: (2, 100)                                                                                            \n",
-       " • Run Command: filter_prime_numbers(2, 100)                                                                       \n",
-       " • Expected Result: 25                                                                                             \n",
-       " • Actual Result: 25                                                                                               \n",
+       " • Test Input: (41, 50)                                                                                            \n",
+       " • Run Command: filter_prime_numbers(41, 50)                                                                       \n",
+       " • Expected Result: 3                                                                                              \n",
+       " • Actual Result: 3                                                                                                \n",
        " • Is Passed: Yes                                                                                                  \n",
        "\n",
        "───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
@@ -268,46 +286,46 @@
        "\n",
        "                                                    \u001b[1mTest Case 0\u001b[0m                                                    \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 2)                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 2)                                                                         \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 1                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 1                                                                                                \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 10)                                                                                             \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 10)                                                                        \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 4                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 4                                                                                                \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mIs Passed:\u001b[0m Yes                                                                                                  \n",
        "\n",
        "\u001b[33m───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\u001b[0m\n",
        "                                                    \u001b[1mTest Case 1\u001b[0m                                                    \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 3)                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 3)                                                                         \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 2                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 2                                                                                                \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (11, 20)                                                                                            \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(11, 20)                                                                       \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 4                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 4                                                                                                \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mIs Passed:\u001b[0m Yes                                                                                                  \n",
        "\n",
        "\u001b[33m───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\u001b[0m\n",
        "                                                    \u001b[1mTest Case 2\u001b[0m                                                    \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 10)                                                                                             \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 10)                                                                        \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 4                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 4                                                                                                \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (21, 30)                                                                                            \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(21, 30)                                                                       \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 2                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 2                                                                                                \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mIs Passed:\u001b[0m Yes                                                                                                  \n",
        "\n",
        "\u001b[33m───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\u001b[0m\n",
        "                                                    \u001b[1mTest Case 3\u001b[0m                                                    \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 20)                                                                                             \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 20)                                                                        \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 8                                                                                              \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 8                                                                                                \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (31, 40)                                                                                            \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(31, 40)                                                                       \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 2                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 2                                                                                                \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mIs Passed:\u001b[0m Yes                                                                                                  \n",
        "\n",
        "\u001b[33m───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\u001b[0m\n",
        "                                                    \u001b[1mTest Case 4\u001b[0m                                                    \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (2, 100)                                                                                            \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(2, 100)                                                                       \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 25                                                                                             \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 25                                                                                               \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mTest Input:\u001b[0m (41, 50)                                                                                            \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mRun Command:\u001b[0m filter_prime_numbers(41, 50)                                                                       \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mExpected Result:\u001b[0m 3                                                                                              \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mActual Result:\u001b[0m 3                                                                                                \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mIs Passed:\u001b[0m Yes                                                                                                  \n",
        "\n",
        "\u001b[33m───────────────────────────────────────────────────────────────────────────────────────────────────────────────────\u001b[0m\n"
@@ -323,7 +341,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -345,7 +363,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
diff --git a/docs/examples/06_skills_refactor.ipynb b/docs/examples/06_skills_refactor.ipynb
index 7c18ed8..44f246b 100644
--- a/docs/examples/06_skills_refactor.ipynb
+++ b/docs/examples/06_skills_refactor.ipynb
@@ -180,7 +180,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "9f7b8af45abf4ef598efa08de48ddc01",
+       "model_id": "8f80adfafef544098b2f5312b9b605b6",
        "version_major": 2,
        "version_minor": 0
       },
@@ -210,6 +210,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/3753f345-\n",
+       "ceed-4928-9099-327d12fa31c8?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=323673;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/3753f345-ceed-4928-9099-327d12fa31c8?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/3753f345-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=323673;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/3753f345-ceed-4928-9099-327d12fa31c8?poll=true\u001b\\\u001b[4;34mceed-4928-9099-327d12fa31c8?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -295,7 +312,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "58dae241e0e5474bb2e974701ddc6021",
+       "model_id": "2465596d3dff439d9fce6d4ab1b46e74",
        "version_major": 2,
        "version_minor": 0
       },
@@ -325,6 +342,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/9b5d62a4-\n",
+       "109d-4d56-a996-a9cbf928c840?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=180123;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/9b5d62a4-109d-4d56-a996-a9cbf928c840?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/9b5d62a4-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=180123;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/9b5d62a4-109d-4d56-a996-a9cbf928c840?poll=true\u001b\\\u001b[4;34m109d-4d56-a996-a9cbf928c840?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -350,7 +384,7 @@
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " cleaned_data, null_count, duplicate_count = data_cleaning_with_stats(input_data, remove_duplicates=True)          \n",
+       " cleaned_data, stats = data_cleaning_with_stats(input_data, remove_duplicates=True)                                \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
@@ -362,8 +396,7 @@
        " • Returns:                                                                                                        \n",
        "    • cleaned_data (array): The cleaned dataset with string 'null'/'NaN' values converted to actual nulls, and     \n",
        "      nulls and duplicates removed based on specified parameters.                                                  \n",
-       "    • null_count (integer): The number of null values removed from the dataset.                                    \n",
-       "    • duplicate_count (integer): The number of duplicate values removed from the dataset.                          \n",
+       "    • stats (dictionary): A dictionary containing the count of null and duplicate values removed from the dataset. \n",
        "
\n"
       ],
       "text/plain": [
@@ -378,7 +411,7 @@
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcleaned_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mnull_count\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mduplicate_count\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_cleaning_with_stats\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mremove_duplicates\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mTrue\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m         \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcleaned_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mstats\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_cleaning_with_stats\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mremove_duplicates\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mTrue\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                               \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
@@ -390,8 +423,7 @@
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mcleaned_data\u001b[0m (array): The cleaned dataset with string 'null'/'NaN' values converted to actual nulls, and     \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0mnulls and duplicates removed based on specified parameters.                                                  \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mnull_count\u001b[0m (integer): The number of null values removed from the dataset.                                    \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mduplicate_count\u001b[0m (integer): The number of duplicate values removed from the dataset.                          \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mstats\u001b[0m (dictionary): A dictionary containing the count of null and duplicate values removed from the dataset. \n"
       ]
      },
      "metadata": {},
@@ -410,7 +442,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "58332991cad240af85d751bd13a5bc0c",
+       "model_id": "787dc04257b046fe96b96b80d630360f",
        "version_major": 2,
        "version_minor": 0
       },
@@ -440,6 +472,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a3c50ed0-\n",
+       "21e1-4d5d-8324-9a247f29c2f7?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=438953;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a3c50ed0-21e1-4d5d-8324-9a247f29c2f7?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a3c50ed0-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=438953;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a3c50ed0-21e1-4d5d-8324-9a247f29c2f7?poll=true\u001b\\\u001b[4;34m21e1-4d5d-8324-9a247f29c2f7?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -812,7 +861,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "6a0e86dd85914422a1d9e70baf3e4119",
+       "model_id": "482c2f1f5ef64aec803292ea4137421a",
        "version_major": 2,
        "version_minor": 0
       },
@@ -842,6 +891,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/02f47717-\n",
+       "dc2e-4908-af17-7403f9732e2d?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=890260;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/02f47717-dc2e-4908-af17-7403f9732e2d?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/02f47717-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=890260;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/02f47717-dc2e-4908-af17-7403f9732e2d?poll=true\u001b\\\u001b[4;34mdc2e-4908-af17-7403f9732e2d?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -859,42 +925,42 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: data_cleaning_and_visualization                                                                           \n",
+       " • Name: clean_and_visualize_data                                                                                  \n",
        " • Description: This skill is responsible for cleaning the input data by removing empty values and then visualizing\n",
-       "   it by generating a bar chart. It provides a simple way to preprocess and understand data distribution and       \n",
-       "   patterns.                                                                                                       \n",
+       "   it by generating a bar chart. It provides a simple way to preprocess and understand data.                       \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " data_cleaning_and_visualization([1, 2, None, 4, 5, None])                                                         \n",
+       " clean_and_visualize_data(input_data)                                                                              \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
        "    • data (array): The input data that needs cleaning and visualization. It should be a list of values.           \n",
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
+       "    • cleaned_data (array): The cleaned data after removing empty values.                                          \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: data_cleaning_and_visualization                                                                           \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: clean_and_visualize_data                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for cleaning the input data by removing empty values and then visualizing\n",
-       "\u001b[1;33m   \u001b[0mit by generating a bar chart. It provides a simple way to preprocess and understand data distribution and       \n",
-       "\u001b[1;33m   \u001b[0mpatterns.                                                                                                       \n",
+       "\u001b[1;33m   \u001b[0mit by generating a bar chart. It provides a simple way to preprocess and understand data.                       \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_cleaning_and_visualization\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m1\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mNone\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m4\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m5\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mNone\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mclean_and_visualize_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                             \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mdata\u001b[0m (array): The input data that needs cleaning and visualization. It should be a list of values.           \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n"
+       "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mcleaned_data\u001b[0m (array): The cleaned data after removing empty values.                                          \n"
       ]
      },
      "metadata": {},
@@ -913,7 +979,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "985e68fa93734802a01ad6ebab432db7",
+       "model_id": "41f38ee4fe9244149b127558f8ed2992",
        "version_major": 2,
        "version_minor": 0
       },
@@ -943,6 +1009,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6c71660e-\n",
+       "f9fc-441c-a399-589ba14ef6d7?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=569225;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6c71660e-f9fc-441c-a399-589ba14ef6d7?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6c71660e-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=569225;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/6c71660e-f9fc-441c-a399-589ba14ef6d7?poll=true\u001b\\\u001b[4;34mf9fc-441c-a399-589ba14ef6d7?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -960,15 +1043,14 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: data_cleaning_visualization_statistics                                                                    \n",
-       " • Description: This skill is responsible for cleaning the input data by removing empty values, visualizing the    \n",
-       "   data by generating a bar chart, and calculating the average value of the data. It provides a comprehensive way  \n",
-       "   to preprocess, understand, and analyze data.                                                                    \n",
+       " • Name: data_analysis                                                                                             \n",
+       " • Description: This skill is responsible for cleaning the input data, visualizing it by generating a bar chart,   \n",
+       "   and calculating its average. It provides a comprehensive way to preprocess, understand, and analyze data.       \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " cleaned_data, average = data_cleaning_visualization_statistics(input_data)                                        \n",
+       " data_analysis(input_data)                                                                                         \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
@@ -977,22 +1059,22 @@
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
        "    • cleaned_data (array): The cleaned data after removing empty values.                                          \n",
-       "    • the_average_value_of (float): The average value of the cleaned data.                                         \n",
+       "    • visualization (object): The visualization of the data.                                                       \n",
+       "    • average (float): The average value of the input data.                                                        \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: data_cleaning_visualization_statistics                                                                    \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for cleaning the input data by removing empty values, visualizing the    \n",
-       "\u001b[1;33m   \u001b[0mdata by generating a bar chart, and calculating the average value of the data. It provides a comprehensive way  \n",
-       "\u001b[1;33m   \u001b[0mto preprocess, understand, and analyze data.                                                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: data_analysis                                                                                             \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for cleaning the input data, visualizing it by generating a bar chart,   \n",
+       "\u001b[1;33m   \u001b[0mand calculating its average. It provides a comprehensive way to preprocess, understand, and analyze data.       \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcleaned_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34maverage\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_cleaning_visualization_statistics\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_analysis\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
@@ -1001,7 +1083,8 @@
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mcleaned_data\u001b[0m (array): The cleaned data after removing empty values.                                          \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mthe_average_value_of\u001b[0m (float): The average value of the cleaned data.                                         \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mvisualization\u001b[0m (object): The visualization of the data.                                                       \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1maverage\u001b[0m (float): The average value of the input data.                                                        \n"
       ]
      },
      "metadata": {},
@@ -1020,7 +1103,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "92d9faa66589490da6353c3930db7e89",
+       "model_id": "b5b2aeed12e648d8bf9151fe5c6bf67f",
        "version_major": 2,
        "version_minor": 0
       },
@@ -1050,6 +1133,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/4e02c38f-\n",
+       "ebda-4e18-949a-bc8667454564?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=469878;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/4e02c38f-ebda-4e18-949a-bc8667454564?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/4e02c38f-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=469878;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/4e02c38f-ebda-4e18-949a-bc8667454564?poll=true\u001b\\\u001b[4;34mebda-4e18-949a-bc8667454564?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -1067,15 +1167,14 @@
        "
\n",
        "                                                  Skill Details:                                                   \n",
        "\n",
-       " • Name: data_cleaning_visualization_statistics                                                                    \n",
-       " • Description: This skill is responsible for cleaning the input data by removing empty values, visualizing the    \n",
-       "   data by generating a bar chart, and calculating the average value of the data. It provides a comprehensive way  \n",
-       "   to preprocess, understand, and analyze data.                                                                    \n",
+       " • Name: data_analysis                                                                                             \n",
+       " • Description: This skill is responsible for cleaning the input data, visualizing it by generating a bar chart,   \n",
+       "   and calculating its average. It provides a comprehensive way to preprocess, understand, and analyze data.       \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " cleaned_data, average = data_cleaning_visualization_statistics(input_data)                                        \n",
+       " data_analysis(input_data)                                                                                         \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
@@ -1084,22 +1183,22 @@
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
        "    • cleaned_data (array): The cleaned data after removing empty values.                                          \n",
-       "    • the_average_value_of (float): The average value of the cleaned data.                                         \n",
+       "    • visualization (object): The visualization of the data.                                                       \n",
+       "    • average (float): The average value of the input data.                                                        \n",
        "
\n"
       ],
       "text/plain": [
        "\n",
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: data_cleaning_visualization_statistics                                                                    \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for cleaning the input data by removing empty values, visualizing the    \n",
-       "\u001b[1;33m   \u001b[0mdata by generating a bar chart, and calculating the average value of the data. It provides a comprehensive way  \n",
-       "\u001b[1;33m   \u001b[0mto preprocess, understand, and analyze data.                                                                    \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: data_analysis                                                                                             \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for cleaning the input data, visualizing it by generating a bar chart,   \n",
+       "\u001b[1;33m   \u001b[0mand calculating its average. It provides a comprehensive way to preprocess, understand, and analyze data.       \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcleaned_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34maverage\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_cleaning_visualization_statistics\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mdata_analysis\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
@@ -1108,7 +1207,8 @@
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mcleaned_data\u001b[0m (array): The cleaned data after removing empty values.                                          \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mthe_average_value_of\u001b[0m (float): The average value of the cleaned data.                                         \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mvisualization\u001b[0m (object): The visualization of the data.                                                       \n",
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1maverage\u001b[0m (float): The average value of the input data.                                                        \n"
       ]
      },
      "metadata": {},
@@ -1176,7 +1276,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "5dcc82dbf2f44092920cd322f1ddb061",
+       "model_id": "2d5d50adfa434150a37aa177f2a14c25",
        "version_major": 2,
        "version_minor": 0
       },
@@ -1206,6 +1306,23 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b0de9ace-\n",
+       "2d39-4f25-80c8-19b8c36ec3bc?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=320681;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b0de9ace-2d39-4f25-80c8-19b8c36ec3bc?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b0de9ace-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=320681;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/b0de9ace-2d39-4f25-80c8-19b8c36ec3bc?poll=true\u001b\\\u001b[4;34m2d39-4f25-80c8-19b8c36ec3bc?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -1244,8 +1361,8 @@
        "                                                  Skill Details:                                                   \n",
        "\n",
        " • Name: visualize_data                                                                                            \n",
-       " • Description: This skill is responsible for visualizing the input data using a bar chart. It provides a          \n",
-       "   comprehensive overview of the dataset.                                                                          \n",
+       " • Description: This skill is responsible for visualizing the input data using a bar chart. It provides a visual   \n",
+       "   overview of the dataset.                                                                                        \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
@@ -1264,8 +1381,8 @@
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: visualize_data                                                                                            \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for visualizing the input data using a bar chart. It provides a          \n",
-       "\u001b[1;33m   \u001b[0mcomprehensive overview of the dataset.                                                                          \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for visualizing the input data using a bar chart. It provides a visual   \n",
+       "\u001b[1;33m   \u001b[0moverview of the dataset.                                                                                        \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
@@ -1289,19 +1406,20 @@
        "                                                  Skill Details:                                                   \n",
        "\n",
        " • Name: calculate_average                                                                                         \n",
-       " • Description: This skill is responsible for calculating the average of the input data.                           \n",
+       " • Description: This skill is responsible for calculating the average of the input data. It provides a statistical \n",
+       "   analysis of the dataset.                                                                                        \n",
        " • Version: 1.0.0                                                                                                  \n",
        " • Usage:                                                                                                          \n",
        "\n",
        "                                                                                                                   \n",
-       " calculate_average(input_data)                                                                                     \n",
+       " average = calculate_average(input_data)                                                                           \n",
        "                                                                                                                   \n",
        "\n",
        " • Parameters:                                                                                                     \n",
        "    • input_data (any): The input dataset to be analyzed.                                                          \n",
        "       • Required: True                                                                                            \n",
        " • Returns:                                                                                                        \n",
-       "    • this_function_return (float): This function returns the average of the input data.                           \n",
+       "    • average (float): The average of the input data.                                                              \n",
        "
\n"
       ],
       "text/plain": [
@@ -1309,19 +1427,20 @@
        "                                                  \u001b[1;4mSkill Details:\u001b[0m                                                   \n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mName\u001b[0m: calculate_average                                                                                         \n",
-       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for calculating the average of the input data.                           \n",
+       "\u001b[1;33m • \u001b[0m\u001b[1mDescription\u001b[0m: This skill is responsible for calculating the average of the input data. It provides a statistical \n",
+       "\u001b[1;33m   \u001b[0manalysis of the dataset.                                                                                        \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mVersion\u001b[0m: 1.0.0                                                                                                  \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mUsage\u001b[0m:                                                                                                          \n",
        "\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcalculate_average\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34maverage\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mcalculate_average\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minput_data\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                          \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
        "\n",
        "\u001b[1;33m • \u001b[0m\u001b[1mParameters\u001b[0m:                                                                                                     \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1minput_data\u001b[0m (any): The input dataset to be analyzed.                                                          \n",
        "\u001b[1;33m   \u001b[0m\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0mRequired: True                                                                                            \n",
        "\u001b[1;33m • \u001b[0m\u001b[1mReturns\u001b[0m:                                                                                                        \n",
-       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1mthis_function_return\u001b[0m (float): This function returns the average of the input data.                           \n"
+       "\u001b[1;33m   \u001b[0m\u001b[1;33m • \u001b[0m\u001b[1maverage\u001b[0m (float): The average of the input data.                                                              \n"
       ]
      },
      "metadata": {},
diff --git a/docs/examples/07_skills_auto_optimize.ipynb b/docs/examples/07_skills_auto_optimize.ipynb
index 075140c..92a3103 100644
--- a/docs/examples/07_skills_auto_optimize.ipynb
+++ b/docs/examples/07_skills_auto_optimize.ipynb
@@ -196,28 +196,106 @@
     {
      "data": {
       "text/html": [
-       "
{\n",
-       "  \"status\": \"success\",\n",
-       "  \"stdout\": \"\",\n",
-       "  \"stderr\": \"\\u001b[33mWARNING: Package(s) not found: itertools\\u001b[0m\\u001b[33m\\n\"\n",
-       "}\n",
+       "
\n",
+       "▌ Installing dependencies                                                                                        \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstalling dependencies\u001b[0m\u001b[35m                                                                                        \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                                                                                                                   \n",
+       " pip show itertools || pip install \"itertools\"                                                                     \n",
+       "                                                                                                                   \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mshow\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mitertools\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m||\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpip\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34minstall\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m\"itertools\"\u001b[0m\u001b[48;2;39;40;34m                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m                                                                                                                   \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
\n",
+       "▌ Install dependencies result: {'status': 'success', 'stdout': '', 'stderr': '\\x1b[33mWARNING: Package(s) not    \n",
+       "▌ found: itertools\\x1b[0m\\x1b[33m\\n\\x1b[0m\\x1b[31mERROR: Could not find a version that satisfies the requirement \n",
+       "▌ itertools (from versions: none)\\x1b[0m\\x1b[31m\\n'}                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[1m{\u001b[0m\n",
-       "  \u001b[1;34m\"status\"\u001b[0m: \u001b[32m\"success\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stdout\"\u001b[0m: \u001b[32m\"\"\u001b[0m,\n",
-       "  \u001b[1;34m\"stderr\"\u001b[0m: \u001b[32m\"\\u001b[33mWARNING: Package(s) not found: itertools\\u001b[0m\\u001b[33m\\n\"\u001b[0m\n",
-       "\u001b[1m}\u001b[0m\n"
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mInstall dependencies result: {'status': 'success', 'stdout': '', 'stderr': '\\x1b[33mWARNING: Package(s) not \u001b[0m\u001b[35m   \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mfound: itertools\\x1b[0m\\x1b[33m\\n\\x1b[0m\\x1b[31mERROR: Could not find a version that satisfies the requirement \u001b[0m\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35mitertools (from versions: none)\\x1b[0m\\x1b[31m\\n'}\u001b[0m\u001b[35m                                                             \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "c772f6a3b2e74b279099a57efef602f6",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Output()"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "725ae410049c4c9c857d330e4e385640",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Output()"
       ]
      },
      "metadata": {},
      "output_type": "display_data"
     },
+    {
+     "data": {
+      "text/html": [
+       "
\n"
+      ],
+      "text/plain": []
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
\n"
+      ],
+      "text/plain": []
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "02dba3e12d8446fe9435a3213dff13d2",
+       "model_id": "a2504e3fc79b43e688dce661b35dd81b",
        "version_major": 2,
        "version_minor": 0
       },
@@ -231,7 +309,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "7ef28624ec87490d934ca62661618de1",
+       "model_id": "0924ad029b6247b3a47cb76c584c16dc",
        "version_major": 2,
        "version_minor": 0
       },
@@ -265,7 +343,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "de9bd6adeea544949701f07d6454a9df",
+       "model_id": "e999daf7a6e1492e8c546b899b741426",
        "version_major": 2,
        "version_minor": 0
       },
@@ -314,7 +392,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "62ed1213deb543d8a0ff18b06e8396ce",
+       "model_id": "44cef59cfedd4730818910c8218ddccb",
        "version_major": 2,
        "version_minor": 0
       },
@@ -344,6 +422,40 @@
      },
      "metadata": {},
      "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/61f07104-\n",
+       "29a1-420c-b831-cbda8df3668e?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=734838;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/61f07104-29a1-420c-b831-cbda8df3668e?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/61f07104-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=734838;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/61f07104-29a1-420c-b831-cbda8df3668e?poll=true\u001b\\\u001b[4;34m29a1-420c-b831-cbda8df3668e?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/7f305c8e-\n",
+       "9bb6-45fc-9035-5af496184866?poll=true                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=99153;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/7f305c8e-9bb6-45fc-9035-5af496184866?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/7f305c8e-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=99153;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/7f305c8e-9bb6-45fc-9035-5af496184866?poll=true\u001b\\\u001b[4;34m9bb6-45fc-9035-5af496184866?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
     }
    ],
    "source": [
@@ -425,9 +537,10 @@
        "                                                                                                                   \n",
        " def solve_game_of_24(numbers):                                                                                    \n",
        "     for permutation in permutations(numbers):                                                                     \n",
+       "         a, b, c, d = permutation                                                                                  \n",
+       "         # Try all possible combinations of arithmetic operations                                                  \n",
        "         for ops in product(['+', '-', '*', '/'], repeat=3):                                                       \n",
-       "             expression = f'(({permutation[0]} {ops[0]} {permutation[1]}) {ops[1]} {permutation[2]}) {ops[2]}      \n",
-       " {permutation[3]}'                                                                                                 \n",
+       "             expression = f'(({a} {ops[0]} {b}) {ops[1]} {c}) {ops[2]} {d}'                                        \n",
        "             try:                                                                                                  \n",
        "                 result = eval(expression)                                                                         \n",
        "                 if result == 24:                                                                                  \n",
@@ -445,9 +558,10 @@
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m                                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mdef\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;166;226;46;48;2;39;40;34msolve_game_of_24\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mnumbers\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m:\u001b[0m\u001b[48;2;39;40;34m                                                                                   \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m    \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mfor\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34min\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutations\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mnumbers\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m:\u001b[0m\u001b[48;2;39;40;34m                                                                    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m        \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34ma\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mb\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mc\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34md\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[48;2;39;40;34m                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m        \u001b[0m\u001b[38;2;149;144;119;48;2;39;40;34m# Try all possible combinations of arithmetic operations\u001b[0m\u001b[48;2;39;40;34m                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m        \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mfor\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34min\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mproduct\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m+\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m-\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m*\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m/\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m,\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mrepeat\u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m3\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m:\u001b[0m\u001b[48;2;39;40;34m                                                      \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m            \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mexpression\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34mf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m((\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m0\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m0\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m1\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m) \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m1\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m) \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[48;2;39;40;34m    \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
-       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mpermutation\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m3\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[48;2;39;40;34m                                                                                                \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
+       "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m            \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mexpression\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34mf\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m((\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34ma\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m0\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mb\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m) \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m1\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mc\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m) \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mops\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m[\u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m2\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m]\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m \u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m{\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34md\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m}\u001b[0m\u001b[38;2;230;219;116;48;2;39;40;34m'\u001b[0m\u001b[48;2;39;40;34m                                       \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m            \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mtry\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m:\u001b[0m\u001b[48;2;39;40;34m                                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m                \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mresult\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m=\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34meval\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m(\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mexpression\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m)\u001b[0m\u001b[48;2;39;40;34m                                                                        \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
        "\u001b[48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m                \u001b[0m\u001b[38;2;102;217;239;48;2;39;40;34mif\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34mresult\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;255;70;137;48;2;39;40;34m==\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m \u001b[0m\u001b[38;2;174;129;255;48;2;39;40;34m24\u001b[0m\u001b[38;2;248;248;242;48;2;39;40;34m:\u001b[0m\u001b[48;2;39;40;34m                                                                                 \u001b[0m\u001b[48;2;39;40;34m \u001b[0m\n",
diff --git a/docs/examples/08_creator_agent.ipynb b/docs/examples/08_creator_agent.ipynb
index 71b1f35..9ef93d5 100644
--- a/docs/examples/08_creator_agent.ipynb
+++ b/docs/examples/08_creator_agent.ipynb
@@ -16,7 +16,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from creator.agents.creator_agent import open_creator_agent"
+    "from creator.agents import create_creator_agent\n",
+    "from creator import config"
    ]
   },
   {
@@ -24,6 +25,15 @@
    "execution_count": 2,
    "metadata": {},
    "outputs": [],
+   "source": [
+    "open_creator_agent = create_creator_agent(config)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
    "source": [
     "messages = [\n",
     "    {\n",
@@ -35,13 +45,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 4,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "6e7dca82c2dd437d943dba009b57d9a9",
+       "model_id": "9f536b3686f94b1dab2dfd584f4851cb",
        "version_major": 2,
        "version_minor": 0
       },
@@ -79,7 +89,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -93,13 +103,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "ba05f0dbad2748379194af32d2818c1d",
+       "model_id": "a59a5d9a43ad4ab59827fbca51396623",
        "version_major": 2,
        "version_minor": 0
       },
@@ -113,7 +123,21 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "d66203c2d2104397837cc61c9763f2dc",
+       "model_id": "1e4ddee0f8614a0d8d5829b3c08261b0",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Output()"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "031a74ab9b284404814fb41b31920d5a",
        "version_major": 2,
        "version_minor": 0
       },
@@ -147,7 +171,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "c6d74112eaa7438ea8dad9bfcfab4764",
+       "model_id": "8c67158044da4b96aa45366119e7ffb1",
        "version_major": 2,
        "version_minor": 0
       },
@@ -161,7 +185,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "28463b69814b46d5bd286b47fe3898cf",
+       "model_id": "185103ec05594133a317a8f4fe1768fa",
        "version_major": 2,
        "version_minor": 0
       },
@@ -195,7 +219,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "abf238be7b554d81bd845596e39e33d1",
+       "model_id": "65425fd5707c40159b5bd51d4fbd412a",
        "version_major": 2,
        "version_minor": 0
       },
@@ -209,7 +233,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "536ad56670e545b785b3aab24246c8dd",
+       "model_id": "9d680c3af9c54b849aaf07b0a9a9a95b",
        "version_major": 2,
        "version_minor": 0
       },
@@ -220,6 +244,16 @@
      "metadata": {},
      "output_type": "display_data"
     },
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "
"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -243,7 +277,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "4ccc78982c0d4cbd906b8d9a0ace7306",
+       "model_id": "262ed48cde63468ca3f695f67131d00d",
        "version_major": 2,
        "version_minor": 0
       },
@@ -277,7 +311,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "5a2e9a34523b4d57ada811f712f751d8",
+       "model_id": "4b74f781084e4861b0d3cffa933762da",
        "version_major": 2,
        "version_minor": 0
       },
@@ -310,13 +344,16 @@
     },
     {
      "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "de0d200823d047e8bd963b6c3321e194",
-       "version_major": 2,
-       "version_minor": 0
-      },
+      "text/html": [
+       "
Langsmith Run URL:                                                                                                 \n",
+       "http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a790ec8f-\n",
+       "71ad-4de2-b81c-74f1575ebb0b?poll=true                                                                              \n",
+       "
\n"
+      ],
       "text/plain": [
-       "Output()"
+       "Langsmith Run URL:                                                                                                 \n",
+       "\u001b]8;id=656281;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a790ec8f-71ad-4de2-b81c-74f1575ebb0b?poll=true\u001b\\\u001b[4;34mhttp://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a790ec8f-\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b]8;id=656281;http://localhost/o/00000000-0000-0000-0000-000000000000/projects/p/856d5024-48fe-4a9c-b1f9-5236f8d3aebd/r/a790ec8f-71ad-4de2-b81c-74f1575ebb0b?poll=true\u001b\\\u001b[4;34m71ad-4de2-b81c-74f1575ebb0b?poll=true\u001b[0m\u001b]8;;\u001b\\                                                                              \n"
       ]
      },
      "metadata": {},
@@ -324,14 +361,13 @@
     },
     {
      "data": {
-      "text/html": [
-       "
\n",
-       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/solve_random_maze                                 \n",
-       "
\n"
-      ],
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "671359bd229c41a3b2c69a96b05faefe",
+       "version_major": 2,
+       "version_minor": 0
+      },
       "text/plain": [
-       "\n",
-       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/solve_random_maze\u001b[0m\u001b[35m                                 \u001b[0m\n"
+       "Output()"
       ]
      },
      "metadata": {},
@@ -340,7 +376,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "be609bd2309e41e99329d856893b39df",
+       "model_id": "d3e46fda387046bc816af54504b016ee",
        "version_major": 2,
        "version_minor": 0
       },
@@ -351,6 +387,21 @@
      "metadata": {},
      "output_type": "display_data"
     },
+    {
+     "data": {
+      "text/html": [
+       "
\n",
+       "▌ saved to /Users/gongjunmin/.cache/open_creator/skill_library/solve_maze                                        \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\n",
+       "\u001b[35m▌ \u001b[0m\u001b[35msaved to /Users/gongjunmin/.cache/open_creator/skill_library/solve_maze\u001b[0m\u001b[35m                                        \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -374,7 +425,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "48a0f6cb1f6d4311bd0f7db3e7dd3994",
+       "model_id": "51ec3eacd34c40688129654e96f9fb31",
        "version_major": 2,
        "version_minor": 0
       },
@@ -412,7 +463,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -426,9 +477,37 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "3f6bd9ad7125466381d2e546d0a76338",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Output()"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "548c2d4a340e4f1ab798cbc2361152e2",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Output()"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/html": [
@@ -558,34 +637,6 @@
      "metadata": {},
      "output_type": "display_data"
     },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "418c6e79ab4047b5bcac823778e0659c",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Output()"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "9b2ce948f375427d9d80e5e0372fec1e",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Output()"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
     {
      "data": {
       "text/html": [
@@ -609,7 +660,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "4b137fec8d4a46b3a1f2fa49f77517da",
+       "model_id": "41340c9f71fa45e3b306ec07eee7ca56",
        "version_major": 2,
        "version_minor": 0
       },

From c07923b41de240b6f04119b2dc7bd2820b22f5f4 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 16:34:46 +0800
Subject: [PATCH 44/56] fix: test subagent run ok

---
 creator/memgpt/agent.py           |  7 +--
 creator/memgpt/memory/__init__.py |  5 +-
 creator/memgpt/message_handler.py |  8 +++-
 creator/memgpt/tools_handler.py   |  2 +-
 docs/configurations.md            | 51 ++++++++++++++++++--
 docs/examples/09_memgpt.ipynb     | 78 +++++++++++++++++++++++++++++++
 6 files changed, 140 insertions(+), 11 deletions(-)
 create mode 100644 docs/examples/09_memgpt.ipynb

diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py
index a41e180..91ce881 100644
--- a/creator/memgpt/agent.py
+++ b/creator/memgpt/agent.py
@@ -52,7 +52,8 @@ def preprocess_inputs(self, inputs: Dict[str, Any]):
         else:
             inputs["session_id"] = self.memory_manager.session_id
 
-        inputs["memory_edit_timestamp"] = self.memory_manager.memory_edit_timestamp
+        # inputs["memory_edit_timestamp"] = self.memory_manager.memory_edit_timestamp
+        inputs["memory_edit_timestamp"] = 0
         inputs["recall_memory_count"] = self.memory_manager.recall_memory_count
         inputs["archival_memory_count"] = self.memory_manager.archival_memory_count
 
@@ -156,7 +157,7 @@ async def arun_workflow(self, inputs: Dict[str, Any], run_manager: Optional[Call
             # construct prompt and run
             prompt = self.construct_prompt(langchain_messages)
             llm_chain = (prompt | llm_with_functions).with_config({"run_name": f"Iteration {counter+1}"})
-            message = await llm_chain.ainvoke(inputs, {"callbacks": run_manager_callbacks})
+            message = llm_chain.invoke(inputs, {"callbacks": run_manager_callbacks})
             await self.memory_manager.add_message(message)
 
             # handle with ai response
@@ -210,7 +211,7 @@ async def _acall(
 
 def create_memgpt(config, subagent=None):
     template = load_system_prompt(config.memgpt_system_prompt_path)
-    function_schemas = load_json_schema(config.memgpt__function_schema_path)
+    function_schemas = load_json_schema(config.memgpt_function_schema_path)
     memory_manager = MemoryManager(config.memgpt_config)
     llm = create_llm(config)
     if subagent is None:
diff --git a/creator/memgpt/memory/__init__.py b/creator/memgpt/memory/__init__.py
index 6cf1953..d15b49e 100644
--- a/creator/memgpt/memory/__init__.py
+++ b/creator/memgpt/memory/__init__.py
@@ -48,7 +48,10 @@ def archival_memory_count(self):
 
     @property
     def is_new_session(self):
-        return len(self.chat_message_history.messages) == 0
+        return len(self.chat_message_history.messages) == 1 and self.chat_message_history.messages[0].type == "human"
+
+    def clear(self):
+        self.chat_message_history.clear()
 
     @property
     def messages(self):
diff --git a/creator/memgpt/message_handler.py b/creator/memgpt/message_handler.py
index 6102294..8b2b093 100644
--- a/creator/memgpt/message_handler.py
+++ b/creator/memgpt/message_handler.py
@@ -49,4 +49,10 @@ def get_initial_boot_messages():
 
 
 def get_login_event():
-    return package_message('login', {"last_login": 'Never (first login)'}, include_location=True)
+    return [
+        {
+            "role": "function",
+            "name": "send_message",
+            "content": package_message('login', {"last_login": 'Never (first login)'}, include_location=True)
+        }
+    ]
diff --git a/creator/memgpt/tools_handler.py b/creator/memgpt/tools_handler.py
index e381a99..227ec28 100644
--- a/creator/memgpt/tools_handler.py
+++ b/creator/memgpt/tools_handler.py
@@ -24,7 +24,7 @@ async def send_message(memgpt, message: str, receiver: str = "human"):
         return
     else:
         # request subagent first
-        messages = memgpt.subagent.run({"messages": {"role": "user", "content": message}})
+        messages = memgpt.subagent.run({"messages": [{"role": "user", "content": message}]})
         for m in messages:
             langchain_message = convert_dict_to_message(m)
             langchain_message.additional_kwargs["subagent"] = memgpt.subagent._chain_type
diff --git a/docs/configurations.md b/docs/configurations.md
index 18f11c7..04cb5b8 100644
--- a/docs/configurations.md
+++ b/docs/configurations.md
@@ -1,17 +1,21 @@
 
 # Configurations
 
+```shell
+creator -c
+```
+
 ```yaml
 LOCAL_SKILL_LIBRARY_PATH: .cache/open_creator/skill_library
 REMOTE_SKILL_LIBRARY_PATH: .cache/open_creator/remote
-LOCAL_SKILL_LIBRARY_VECTORD_PATH: .cache/open_creator/vectordb/
 PROMPT_CACHE_HISTORY_PATH: .cache/open_creator/prompt_cache/
+VECTORD_PATH: .cache/open_creator/vectordb/
 LOGGER_CACHE_PATH: .cache/open_creator/logs/
-SKILL_EXTRACT_AGENT_CACHE_PATH: .cache/open_creator/llm_cache
+LLM_CACHE_PATH: .cache/open_creator/llm_cache
+EMBEDDING_CACHE_PATH: .cache/open_creator/embeddings/
 OFFICIAL_SKILL_LIBRARY_PATH: timedomain/skill-library
 OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH: timedomain/skill-library-template
 
-BUILD_IN_SKILL_LIBRARY_DIR: skill_library/open-creator/
 
 # for AZURE, it is your_deployment_id
 # for ANTHROPIC, it is claude-2
@@ -19,7 +23,7 @@ BUILD_IN_SKILL_LIBRARY_DIR: skill_library/open-creator/
 # for huggingface, it is huggingface/WizardLM/WizardCoder-Python-34B-V1.0 model path
 # for ollama, it is like ollama/llama2
 # the default is openai/gpt-3.5
-MODEL_NAME: gpt-3.5-turbo-16k
+MODEL_NAME: gpt-4
 TEMPERATURE: 0 # only 0 can use llm_cache
 
 USE_AZURE: false
@@ -37,4 +41,41 @@ VERTEX_LOCATION: ""
 
 HUGGINGFACE_API_KEY: ""
 HUGGINGFACE_API_BASE: ""
-```
\ No newline at end of file
+
+# for langsmith trace
+LANGCHAIN_ENDPOINT: 
+LANGCHAIN_API_KEY: 
+LANGCHAIN_TRACING_V2: true
+LANGCHAIN_PROJECT: "open-creator"
+
+# for memgpt
+MEMGPT_CONFIG: 
+  MEMORY_PATH: .cache/open_creator/memory
+  PERSONA: |
+    The following is a blank slate starter persona, I need to expand this to develop my own personality.
+
+    My name is MemGPT.
+    I am kind, thoughtful, and inquisitive.
+
+  HUMAN: |
+    This is what I know so far about the user, I should expand this as I learn more about them.
+
+    First name: Chad
+    Last name: ?
+    Gender: Male
+    Age: ?
+    Nationality: ?
+    Occupation: Computer science PhD student at UC Berkeley
+    Interests: Formula 1, Sailing, Taste of the Himalayas Restaurant in Berkeley, CSGO
+  
+  AGENT_SUBTASKS: |
+      - create/save/search skill
+      - run/test/refactor skill
+      - show skill
+  
+  SUMMARY_WARNING_TOKENS: 6000
+  CORE_MEMORY_PERSONA_CHAR_LIMIT: 2000
+  CORE_MEMORY_HUMAN_CHAR_LIMIT: 2000
+  PAGE_SIZE: 5
+  USE_VECTOR_SEARCH: true
+```
diff --git a/docs/examples/09_memgpt.ipynb b/docs/examples/09_memgpt.ipynb
new file mode 100644
index 0000000..20c6fc9
--- /dev/null
+++ b/docs/examples/09_memgpt.ipynb
@@ -0,0 +1,78 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from creator import config\n",
+    "from creator.memgpt import create_memgpt"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "session_id = \"test_session\"\n",
+    "config.memgpt_config.session_id = session_id\n",
+    "memgpt = create_memgpt(config)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "memgpt.memory_manager.clear()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "memgpt.memory_manager.messages"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "\n",
+    "while 1:\n",
+    "    user_request = input(\"> \")\n",
+    "    if user_request.startswith(\"/exit\"):\n",
+    "        break\n",
+    "    session_id = await memgpt.arun({\"user_request\": user_request, \"session_id\": session_id})"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "open_creator_online",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.0"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

From 7af161b4f6b1bb527ceed26fb034de01c5573c1a Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 16:45:36 +0800
Subject: [PATCH 45/56] fix: typo

---
 creator/client/repl/handler.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/creator/client/repl/handler.py b/creator/client/repl/handler.py
index e91be5f..9b1a15d 100644
--- a/creator/client/repl/handler.py
+++ b/creator/client/repl/handler.py
@@ -191,7 +191,7 @@ async def agent_handler(self, request):
         inputs = {"messages": messages, "verbose": True}
         with self.console.status("[blue]Thinking[/blue]", spinner="circleHalves"):
             if self.interpreter:
-                messages = self.code_interpreter_agent.run(inputs)
+                messages = self.interpreter_agent.run(inputs)
             else:
                 messages = self.open_creator_agent.run(inputs)
         output = "\n".join([self.convert_agent_message(message) for message in messages[len(self.messages)+1:]])

From 7670391972a2aa9915cc94eb9c4dd2bac509ff0b Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 17:23:34 +0800
Subject: [PATCH 46/56] fix: subagent content, total_tries, set config update
 env

---
 creator/config.yaml               | 2 +-
 creator/memgpt/agent.py           | 3 ++-
 creator/memgpt/message_handler.py | 2 +-
 creator/memgpt/tools_handler.py   | 2 +-
 creator/utils/attri_dict.py       | 7 +++++++
 5 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/creator/config.yaml b/creator/config.yaml
index eb4e8e5..1ca3bc6 100644
--- a/creator/config.yaml
+++ b/creator/config.yaml
@@ -37,7 +37,7 @@ HUGGINGFACE_API_BASE: ""
 # for langsmith trace
 LANGCHAIN_ENDPOINT: 
 LANGCHAIN_API_KEY: 
-LANGCHAIN_TRACING_V2: true
+LANGCHAIN_TRACING_V2: false
 LANGCHAIN_PROJECT: "open-creator"
 
 # for memgpt
diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py
index 91ce881..7f5f2eb 100644
--- a/creator/memgpt/agent.py
+++ b/creator/memgpt/agent.py
@@ -27,6 +27,7 @@
 
 
 class MemGPT(BaseAgent):
+    total_tries: 5
     subagent: Optional[BaseAgent] = None
     pause_heartbeats_start: Optional[datetime.datetime] = None
     function_failed: bool = False
@@ -143,7 +144,7 @@ async def arun_workflow(self, inputs: Dict[str, Any], run_manager: Optional[Call
         user_message = package_message(message_type="user_message", extra_info={"message": user_request})
 
         counter = 0
-        while True:
+        while True or counter < self.total_tries:
             self.start_callbacks()
             skil_next_user_input = False
 
diff --git a/creator/memgpt/message_handler.py b/creator/memgpt/message_handler.py
index 8b2b093..3075325 100644
--- a/creator/memgpt/message_handler.py
+++ b/creator/memgpt/message_handler.py
@@ -25,7 +25,7 @@ def package_message(message_type=None, extra_info=None, include_location=False,
     if extra_info:
         packaged_message.update(extra_info)
 
-    return json.dumps(packaged_message)
+    return json.dumps(packaged_message, ensure_ascii=False)
 
 
 def get_initial_boot_messages():
diff --git a/creator/memgpt/tools_handler.py b/creator/memgpt/tools_handler.py
index 227ec28..2061459 100644
--- a/creator/memgpt/tools_handler.py
+++ b/creator/memgpt/tools_handler.py
@@ -30,7 +30,7 @@ async def send_message(memgpt, message: str, receiver: str = "human"):
             langchain_message.additional_kwargs["subagent"] = memgpt.subagent._chain_type
             memgpt.memory_manager.add_message(message)
         last_m = messages[-1]
-        return last_m.content
+        return last_m.get("content", "No Output")
 
 
 async def pause_heartbeats(memgpt, minutes: int, max_pause: int = 360):
diff --git a/creator/utils/attri_dict.py b/creator/utils/attri_dict.py
index 0ac26ff..71bb396 100644
--- a/creator/utils/attri_dict.py
+++ b/creator/utils/attri_dict.py
@@ -1,3 +1,6 @@
+import os
+
+
 class AttrDict(dict):
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
@@ -15,6 +18,10 @@ def __setattr__(self, key, value):
         if isinstance(value, dict):
             value = AttrDict(value)
         self[key] = value
+        if isinstance(value, (str, int, float)):
+            os.environ[key] = value
+        elif isinstance(value, bool):
+            os.environ[key] = str(value).lower()
 
     def __delattr__(self, key):
         if key in self:

From 2ab4fcdb31d2fabe5ed2ebc707a82a7644d251ad Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 18:20:16 +0800
Subject: [PATCH 47/56] fix: check langsmith ok

---
 creator/utils/langsmith_utils.py | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/creator/utils/langsmith_utils.py b/creator/utils/langsmith_utils.py
index 4f60508..0c15dbc 100644
--- a/creator/utils/langsmith_utils.py
+++ b/creator/utils/langsmith_utils.py
@@ -1,21 +1,22 @@
+import os
 from langsmith import Client
-from langsmith.utils import LangSmithError
-from .printer import print
+from langsmith.utils import LangSmithConnectionError
+from printer import print
 
 
 def check_langsmith_ok():
     cli = Client()
-    ok = False
+    if os.environ.get("LANGCHAIN_TRACING_V2", "false") == "false":
+        return False
     try:
         cli.read_project(project_name="open-creator")
-    except LangSmithError as e:
-        if "Bad Gateway" in str(e):
+    except LangSmithConnectionError as e:
+        if "Connection error" in str(e):
             print("[red]Warning:[/red] [yellow]Langsmith is not running. Please run `langsmith start`.[/yellow]")
-            return ok
+            return False
         else:
             cli.create_project(project_name="open-creator")
-    ok = True
-    return ok
+    return True
 
 
 langsmith_ok = check_langsmith_ok()

From 1b044bffd056901259bb5ab48823c72720cd234e Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 18:31:38 +0800
Subject: [PATCH 48/56] fix: typo

---
 creator/utils/langsmith_utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/creator/utils/langsmith_utils.py b/creator/utils/langsmith_utils.py
index 0c15dbc..e2f406e 100644
--- a/creator/utils/langsmith_utils.py
+++ b/creator/utils/langsmith_utils.py
@@ -1,7 +1,7 @@
 import os
 from langsmith import Client
 from langsmith.utils import LangSmithConnectionError
-from printer import print
+from .printer import print
 
 
 def check_langsmith_ok():

From f0b3d5f1c9a7ce5a2f6331e588ee2771389f6979 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 18:42:18 +0800
Subject: [PATCH 49/56] fix: fix python shell code startswith '!'

---
 creator/code_interpreter/__init__.py | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/creator/code_interpreter/__init__.py b/creator/code_interpreter/__init__.py
index 51aca45..0b525c7 100644
--- a/creator/code_interpreter/__init__.py
+++ b/creator/code_interpreter/__init__.py
@@ -66,6 +66,13 @@ def clean_code(self, code: str) -> str:
         # replace tab to 4
         return code
 
+    def hot_fix(self, language, code):
+        # fix python code start with !
+        if language == "python" and code.startswith("!"):
+            language = "shell"
+            code = code[1:]
+        return language, code
+
     def _run(
         self,
         language: str,
@@ -76,6 +83,7 @@ def _run(
         language = language.lower()
         if language not in language_map:
             return {"status": "error", "stdout": "", "stderr": f"Language {language} not supported, Only support {list(language_map.keys())}"}
+        language, code = self.hot_fix(language, code)
         if language not in self.interpreters:
             self.add_interpreter(language=language)
         code = self.clean_code(code)

From acdc11a1153284358c10b807aa1d97433fadc7ac Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 18:52:35 +0800
Subject: [PATCH 50/56] fix: typo error

---
 creator/memgpt/agent.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py
index 7f5f2eb..48ab354 100644
--- a/creator/memgpt/agent.py
+++ b/creator/memgpt/agent.py
@@ -27,7 +27,7 @@
 
 
 class MemGPT(BaseAgent):
-    total_tries: 5
+    total_tries: int = 5
     subagent: Optional[BaseAgent] = None
     pause_heartbeats_start: Optional[datetime.datetime] = None
     function_failed: bool = False

From 45df03985fd1f62124167bf61091c70e0c4d3df8 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Tue, 31 Oct 2023 19:29:19 +0800
Subject: [PATCH 51/56] fix: no pass

---
 creator/llm/chatllm_with_trim.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/creator/llm/chatllm_with_trim.py b/creator/llm/chatllm_with_trim.py
index 94cbde9..ab1d63e 100644
--- a/creator/llm/chatllm_with_trim.py
+++ b/creator/llm/chatllm_with_trim.py
@@ -17,11 +17,9 @@ class ChatOpenAIWithTrim(TrimMixin, ChatOpenAI):
     cache: bool = True
     function_calls: Optional[List[Dict]] = None
     trimed: bool = False
-    pass
 
 
 class AzureChatOpenAIWithTrim(TrimMixin, AzureChatOpenAI):
     cache: bool = True
     function_calls: Optional[List[Dict]] = None
     trimed: bool = False
-    pass

From 50b222f79f906e7aacefbdc0007bacf65f94f063 Mon Sep 17 00:00:00 2001
From: Sayo 
Date: Tue, 31 Oct 2023 22:24:21 +0800
Subject: [PATCH 52/56] [feat] config agent model

---
 creator/agents/creator_agent.py         | 2 +-
 creator/agents/extractor_agent.py       | 2 +-
 creator/agents/interpreter_agent.py     | 2 +-
 creator/agents/prompt_enhancer_agent.py | 2 +-
 creator/agents/refactor_agent.py        | 2 +-
 creator/agents/tester_agent.py          | 2 +-
 creator/config.yaml                     | 9 +++++++++
 creator/config/library.py               | 4 ++++
 creator/llm/llm_creator.py              | 7 +++++--
 creator/memgpt/agent.py                 | 2 +-
 10 files changed, 25 insertions(+), 9 deletions(-)

diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py
index c2167f7..7791289 100644
--- a/creator/agents/creator_agent.py
+++ b/creator/agents/creator_agent.py
@@ -72,7 +72,7 @@ def create_creator_agent(config):
     code_interpreter = SafePythonInterpreter(allowed_functions=ALLOWED_FUNCTIONS, allowed_methods=ALLOW_METHODS, redirect_output=True)
     code_interpreter.setup(IMPORT_CODE)
     chain = CreatorAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.CREATOR_AGENT),
         system_template=template,
         tools=[code_interpreter],
         function_schemas=[code_interpreter.to_function_schema()],
diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py
index 1d8b87d..228a014 100644
--- a/creator/agents/extractor_agent.py
+++ b/creator/agents/extractor_agent.py
@@ -51,7 +51,7 @@ def create_skill_extractor_agent(config):
     }
 
     chain = SkillExtractorAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.EXTRACTOR_AGENT),
         system_template=template,
         function_schemas=[function_schema],
         verbose=False
diff --git a/creator/agents/interpreter_agent.py b/creator/agents/interpreter_agent.py
index 68c63c7..c03c7c7 100644
--- a/creator/agents/interpreter_agent.py
+++ b/creator/agents/interpreter_agent.py
@@ -54,7 +54,7 @@ def create_code_interpreter_agent(config):
     function_schema = tool.to_function_schema()
     template = load_system_prompt(config.interpreter_agent_prompt_path)
     chain = CodeInterpreterAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.INTERPRETER_AGENT),
         system_template=template,
         function_schemas=[function_schema],
         tools=[tool],
diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py
index 1434d2d..4faa7fc 100644
--- a/creator/agents/prompt_enhancer_agent.py
+++ b/creator/agents/prompt_enhancer_agent.py
@@ -40,7 +40,7 @@ def create_prompt_enhancer_agent(config):
     template = load_system_prompt(config.prompt_enhancer_agent_prompt_path)
     function_schema = load_json_schema(config.prompt_enhancer_schema_path)
     chain = PromptEnhancerAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.PROMPT_ENHANCER_AGENT),
         system_template=template,
         function_schemas=[function_schema],
         verbose=False
diff --git a/creator/agents/refactor_agent.py b/creator/agents/refactor_agent.py
index 7b0b479..cdcb2b2 100644
--- a/creator/agents/refactor_agent.py
+++ b/creator/agents/refactor_agent.py
@@ -58,7 +58,7 @@ def create_code_refactor_agent(config):
     }
 
     chain = CodeRefactorAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.REFACTOR_AGENT),
         system_template=template,
         function_schemas=[function_schema],
         verbose=False
diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py
index 3e8d194..f80b2ff 100644
--- a/creator/agents/tester_agent.py
+++ b/creator/agents/tester_agent.py
@@ -73,7 +73,7 @@ def create_code_tester_agent(config):
     code_interpreter_function_schema = tool.to_function_schema()
     test_summary_function_schema = load_json_schema(config.testsummary_function_schema_path)
     chain = CodeTesterAgent(
-        llm=create_llm(config),
+        llm=create_llm(config, config.agent_model_config.TESTER_AGENT),
         system_template=template,
         function_schemas=[code_interpreter_function_schema, test_summary_function_schema],
         tools=[tool],
diff --git a/creator/config.yaml b/creator/config.yaml
index 1ca3bc6..44f8bdf 100644
--- a/creator/config.yaml
+++ b/creator/config.yaml
@@ -70,3 +70,12 @@ MEMGPT_CONFIG:
   CORE_MEMORY_HUMAN_CHAR_LIMIT: 2000
   PAGE_SIZE: 5
   USE_VECTOR_SEARCH: true
+
+AGENT_MODEL_CONFIG:
+  CREATOR_AGENT: gpt-4
+  INTERPRETER_AGENT: gpt-4
+  TESTER_AGENT: gpt-4
+  MEMGPT: gpt-4
+  EXTRACTOR_AGENT: gpt-3.5-turbo-16k
+  PROMPT_ENHANCER_AGENT: gpt-3.5-turbo
+  REFACTOR_AGENT: gpt-3.5-turbo-16k
diff --git a/creator/config/library.py b/creator/config/library.py
index ee88a10..a3c60ca 100644
--- a/creator/config/library.py
+++ b/creator/config/library.py
@@ -34,6 +34,8 @@ def resolve_path(path):
 
 memgpt_config = yaml_config.MEMGPT_CONFIG
 
+angent_model_config = yaml_config.AGENT_MODEL_CONFIG
+
 _official_skill_library_path = resolve_path(yaml_config.OFFICIAL_SKILL_LIBRARY_PATH)
 _official_skill_library_template_path = resolve_path(yaml_config.OFFICIAL_SKILL_LIBRARY_TEMPLATE_PATH)
 
@@ -92,6 +94,7 @@ class LibraryConfig(BaseModel):
     memgpt_function_schema_path: str = os.path.join(project_dir, "prompts", "memgpt_function_schema.json")
 
     memgpt_config: dict = None
+    agent_model_config: dict = None
 
     use_rich: bool = True
     use_file_logger: bool = False
@@ -99,5 +102,6 @@ class LibraryConfig(BaseModel):
 
 config = LibraryConfig()
 config.memgpt_config = memgpt_config
+config.agent_model_config = angent_model_config
 
 set_llm_cache(SQLiteCache(database_path=f"{config.llm_cache_path}/.langchain.db"))
diff --git a/creator/llm/llm_creator.py b/creator/llm/llm_creator.py
index 3de5513..8b9de16 100644
--- a/creator/llm/llm_creator.py
+++ b/creator/llm/llm_creator.py
@@ -4,10 +4,13 @@
 from .chatllm_with_trim import ChatOpenAIWithTrim, AzureChatOpenAIWithTrim
 
 
-def create_llm(config):
+def create_llm(config, model=None):
     use_azure = True if os.getenv("OPENAI_API_TYPE", None) == "azure" else False
 
-    model_name = config.model
+    if model is None or model == "":
+        model_name = config.model
+    else:
+        model_name = model
     temperature = config.temperature
     streaming = config.use_stream_callback
     callbacks = [OutputBufferStreamingHandler()]
diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py
index 7f5f2eb..b83862c 100644
--- a/creator/memgpt/agent.py
+++ b/creator/memgpt/agent.py
@@ -214,7 +214,7 @@ def create_memgpt(config, subagent=None):
     template = load_system_prompt(config.memgpt_system_prompt_path)
     function_schemas = load_json_schema(config.memgpt_function_schema_path)
     memory_manager = MemoryManager(config.memgpt_config)
-    llm = create_llm(config)
+    llm = create_llm(config, config.agent_model_config.MEMGPT)
     if subagent is None:
         subagent = create_creator_agent(config)
     chain = MemGPT(

From 9fcb7304ab859fd15851fea66ca7c155f5cb5e37 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Wed, 1 Nov 2023 11:49:18 +0800
Subject: [PATCH 53/56] refactor: BaseStreamingHandler and dummy chunk message

---
 creator/callbacks/streaming_stdout.py | 80 +++++++--------------------
 1 file changed, 21 insertions(+), 59 deletions(-)

diff --git a/creator/callbacks/streaming_stdout.py b/creator/callbacks/streaming_stdout.py
index bf8663b..fdd9f22 100644
--- a/creator/callbacks/streaming_stdout.py
+++ b/creator/callbacks/streaming_stdout.py
@@ -1,88 +1,50 @@
-from typing import Any, Dict, List
+from typing import Any, Callable
 
 from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
+from langchain.schema.messages import AIMessageChunk
 
 from .buffer_manager import buffer_output_manager
 from .rich_manager import rich_output_manager
 from .file_manager import file_output_manager
 
 
-class OutputBufferStreamingHandler(StreamingStdOutCallbackHandler):
+class BaseStreamingHandler(StreamingStdOutCallbackHandler):
+    def __init__(self, output_manager: Callable):
+        self.output_manager = output_manager
 
     def on_chain_start(self, **kwargs: Any) -> None:
-        """Run when chain starts running."""
         agent_name = kwargs.get("agent_name")
-        buffer_output_manager.add(agent_name)
+        self.output_manager.add(agent_name)
 
     def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
-        """Run on new LLM token. Only available when streaming is enabled."""
         chunk = kwargs.get("chunk", None)
-        if chunk is not None:
-            buffer_output_manager.update(chunk)
+        if chunk is None:
+            chunk = AIMessageChunk(content=token, additional_kwargs=kwargs)
+        self.output_manager.update(chunk)
 
     def on_tool_end(self, **kwargs: Any) -> Any:
         chunk = kwargs.get("chunk", None)
         if chunk is not None:
-            buffer_output_manager.update_tool_result(chunk)
+            self.output_manager.update_tool_result(chunk)
 
     def on_chain_error(self, error: BaseException, **kwargs: Any) -> None:
-        buffer_output_manager.finish(err=error)
+        self.output_manager.finish(err=error)
 
     def on_chain_end(self, **kwargs: Any) -> None:
-        """Run when chain finishes running."""
         message = kwargs.get("message", None)
-        buffer_output_manager.finish(message=message)
-
-
-class RichTerminalStreamingHandler(StreamingStdOutCallbackHandler):
-
-    def on_chain_start(self, **kwargs: Any) -> None:
-        """Run when chain starts running."""
-        agent_name = kwargs.get("agent_name")
-        rich_output_manager.add(agent_name)
-
-    def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
-        """Run on new LLM token. Only available when streaming is enabled."""
-        chunk = kwargs.get("chunk", None)
-        if chunk is not None:
-            rich_output_manager.update(chunk)
+        self.output_manager.finish(message=message)
 
-    def on_tool_end(self, **kwargs: Any) -> Any:
-        chunk = kwargs.get("chunk", None)
-        if chunk is not None:
-            rich_output_manager.update_tool_result(chunk)
 
-    def on_chain_error(self, error: BaseException, **kwargs: Any) -> None:
-        rich_output_manager.finish(err=error)
+class OutputBufferStreamingHandler(BaseStreamingHandler):
+    def __init__(self):
+        super().__init__(buffer_output_manager)
 
-    def on_chain_end(self, **kwargs: Any) -> None:
-        """Run when chain finishes running."""
-        message = kwargs.get("message", None)
-        rich_output_manager.finish(message=message)
 
+class RichTerminalStreamingHandler(BaseStreamingHandler):
+    def __init__(self):
+        super().__init__(rich_output_manager)
 
-class FileLoggerStreamingHandler(StreamingStdOutCallbackHandler):
 
-    def on_chain_start(self, **kwargs: Any) -> None:
-        """Run when chain starts running."""
-        agent_name = kwargs.get("agent_name")
-        file_output_manager.add(agent_name)
-
-    def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
-        """Run on new LLM token. Only available when streaming is enabled."""
-        chunk = kwargs.get("chunk", None)
-        if chunk is not None:
-            file_output_manager.update(chunk)
-
-    def on_tool_end(self, **kwargs: Any) -> Any:
-        chunk = kwargs.get("chunk", None)
-        if chunk is not None:
-            file_output_manager.update_tool_result(chunk)
-
-    def on_chain_error(self, error: BaseException, **kwargs: Any) -> None:
-        file_output_manager.finish(err=error)
-
-    def on_chain_end(self, **kwargs: Any) -> None:
-        """Run when chain finishes running."""
-        message = kwargs.get("message", None)
-        file_output_manager.finish(message=message)
+class FileLoggerStreamingHandler(BaseStreamingHandler):
+    def __init__(self):
+        super().__init__(file_output_manager)

From 857d586e89d0fd55e12a10e93b67046cfda01213 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Wed, 1 Nov 2023 15:26:49 +0800
Subject: [PATCH 54/56] fix: streamlit subprocess not stop

---
 creator/app/streamlit_app.py |  5 ++++-
 creator/client/command.py    | 16 ++++++++++++++--
 2 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/creator/app/streamlit_app.py b/creator/app/streamlit_app.py
index d41fdc9..5296913 100644
--- a/creator/app/streamlit_app.py
+++ b/creator/app/streamlit_app.py
@@ -37,11 +37,14 @@ def setup_slidebar():
             os.environ["OPENAI_API_KEY"] = openai_api_key
         model_list = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "gpt-4"]
         model = st.selectbox("Model", model_list, key="model")
-        config.model = model
         temperature = st.slider("Temperature", 0.0, 1.0, 0.0, 0.05, key="temperature")
         config.temperature = temperature
         agent_list = ["creator_agent", "interpreter_agent"]
         selected_agent = st.selectbox("Agent", agent_list, key="agent")
+        if selected_agent == "creator_agent":
+            config.agent_model_config.CREATOR_AGENT = model
+        else:
+            config.agent_model_config.INTERPRETER_AGENT = model
         if agent_name != selected_agent:
             agent_name = selected_agent
             add_session()
diff --git a/creator/client/command.py b/creator/client/command.py
index 200296d..9826f7d 100644
--- a/creator/client/command.py
+++ b/creator/client/command.py
@@ -267,11 +267,23 @@ def cmd_client():
 
     if args.command == "ui":
         import os
+        import sys
+        import atexit
         import subprocess
         streamlit_app_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "../app", "streamlit_app.py")
         env = os.environ.copy()
-        subprocess.Popen(["streamlit", "run", streamlit_app_path], env=env)
-        return
+        process = subprocess.Popen(["streamlit", "run", streamlit_app_path], env=env)
+        
+        def terminate_process(process):
+            if process:
+                process.terminate()
+
+        atexit.register(terminate_process, process)
+        try:
+            process.wait()
+        except KeyboardInterrupt:
+            pass
+        sys.exit(process.returncode)
 
 
 if __name__ == "__main__":

From d3d3ad0a4f4b11d04d33740286627e2bca7eef1f Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Wed, 1 Nov 2023 16:01:21 +0800
Subject: [PATCH 55/56] refactor: mv code from init to manager.py

---
 creator/memgpt/memory/__init__.py | 83 ++-----------------------------
 creator/memgpt/memory/manager.py  | 81 ++++++++++++++++++++++++++++++
 2 files changed, 85 insertions(+), 79 deletions(-)
 create mode 100644 creator/memgpt/memory/manager.py

diff --git a/creator/memgpt/memory/__init__.py b/creator/memgpt/memory/__init__.py
index d15b49e..2dbb419 100644
--- a/creator/memgpt/memory/__init__.py
+++ b/creator/memgpt/memory/__init__.py
@@ -1,81 +1,6 @@
-from .core_memory import CoreMemory
-from .recall_memory import RecallMemory
-from .archival_memory import ArchivalMemory
-from .builder import build_memory
-from ..time_utils import get_local_time
+from .manager import MemoryManager
 
 
-class MemoryManager:
-
-    def __init__(self, memory_config):
-        self.chat_message_history = build_memory(memory_path=memory_config.MEMORY_PATH, session_id=memory_config.session_id)
-        self.session_id = self.chat_message_history.session_id
-        self.core_memory = CoreMemory(
-            persona=memory_config.PERSONA,
-            human=memory_config.HUMAN,
-            persona_char_limit=memory_config.CORE_MEMORY_PERSONA_CHAR_LIMIT,
-            human_char_limit=memory_config.CORE_MEMORY_HUMAN_CHAR_LIMIT,
-        )
-        self.recall_memory = RecallMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH)
-        self.archival_memory = ArchivalMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH)
-        self.page_size = self.recall_memory.page_size = self.archival_memory.page_size = memory_config.PAGE_SIZE
-
-    @property
-    def human(self):
-        return self.core_memory.human
-
-    @property
-    def persona(self):
-        return self.core_memory.persona
-
-    @property
-    def memory_edit_timestamp(self):
-        messages = self.chat_message_history.messages
-        now = get_local_time()
-        if messages:
-            return messages[-1].additional_kwargs.get("created_at", now)
-        return now
-
-    @property
-    def recall_memory_count(self):
-        messages = self.chat_message_history.messages
-        return len([m for m in messages if m.type != "archival"])
-
-    @property
-    def archival_memory_count(self):
-        messages = self.chat_message_history.messages
-        return len([m for m in messages if m.type == "archival"])
-
-    @property
-    def is_new_session(self):
-        return len(self.chat_message_history.messages) == 1 and self.chat_message_history.messages[0].type == "human"
-
-    def clear(self):
-        self.chat_message_history.clear()
-
-    @property
-    def messages(self):
-        messages = self.chat_message_history.messages
-        # filter archival and subagent
-        messages = [m for m in messages if m.type != "archival" and "subagent" not in m.additional_kwargs]
-        return messages
-
-    async def add_user_message(self, message):
-        self.chat_message_history.add_user_message(message)
-
-    async def add_message(self, message):
-        self.chat_message_history.add_message(message)
-
-    def get_memory(self, memory_type):
-        memory = getattr(self, f"{memory_type}_memory", None)
-        assert memory is not None, f"Memory type {memory_type} not found"
-        return memory
-
-    async def add(self, memory_type, message, name=None):
-        return await self.get_memory(memory_type=memory_type).add(message, name=name)
-
-    async def modify(self, memory_type, old_content, new_content, name=None):
-        return await self.get_memory(memory_type=memory_type).modify(old_content=old_content, new_content=new_content, name=name)
-
-    async def search(self, memory_type, query, page, start_date=None, end_date=None):
-        return await self.get_memory(memory_type=memory_type).search(query=query, page=page, start_date=start_date, end_date=end_date)
+__all__ = [
+    "MemoryManager"
+]
diff --git a/creator/memgpt/memory/manager.py b/creator/memgpt/memory/manager.py
new file mode 100644
index 0000000..d15b49e
--- /dev/null
+++ b/creator/memgpt/memory/manager.py
@@ -0,0 +1,81 @@
+from .core_memory import CoreMemory
+from .recall_memory import RecallMemory
+from .archival_memory import ArchivalMemory
+from .builder import build_memory
+from ..time_utils import get_local_time
+
+
+class MemoryManager:
+
+    def __init__(self, memory_config):
+        self.chat_message_history = build_memory(memory_path=memory_config.MEMORY_PATH, session_id=memory_config.session_id)
+        self.session_id = self.chat_message_history.session_id
+        self.core_memory = CoreMemory(
+            persona=memory_config.PERSONA,
+            human=memory_config.HUMAN,
+            persona_char_limit=memory_config.CORE_MEMORY_PERSONA_CHAR_LIMIT,
+            human_char_limit=memory_config.CORE_MEMORY_HUMAN_CHAR_LIMIT,
+        )
+        self.recall_memory = RecallMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH)
+        self.archival_memory = ArchivalMemory(message_database=self.chat_message_history, use_vector_search=memory_config.USE_VECTOR_SEARCH)
+        self.page_size = self.recall_memory.page_size = self.archival_memory.page_size = memory_config.PAGE_SIZE
+
+    @property
+    def human(self):
+        return self.core_memory.human
+
+    @property
+    def persona(self):
+        return self.core_memory.persona
+
+    @property
+    def memory_edit_timestamp(self):
+        messages = self.chat_message_history.messages
+        now = get_local_time()
+        if messages:
+            return messages[-1].additional_kwargs.get("created_at", now)
+        return now
+
+    @property
+    def recall_memory_count(self):
+        messages = self.chat_message_history.messages
+        return len([m for m in messages if m.type != "archival"])
+
+    @property
+    def archival_memory_count(self):
+        messages = self.chat_message_history.messages
+        return len([m for m in messages if m.type == "archival"])
+
+    @property
+    def is_new_session(self):
+        return len(self.chat_message_history.messages) == 1 and self.chat_message_history.messages[0].type == "human"
+
+    def clear(self):
+        self.chat_message_history.clear()
+
+    @property
+    def messages(self):
+        messages = self.chat_message_history.messages
+        # filter archival and subagent
+        messages = [m for m in messages if m.type != "archival" and "subagent" not in m.additional_kwargs]
+        return messages
+
+    async def add_user_message(self, message):
+        self.chat_message_history.add_user_message(message)
+
+    async def add_message(self, message):
+        self.chat_message_history.add_message(message)
+
+    def get_memory(self, memory_type):
+        memory = getattr(self, f"{memory_type}_memory", None)
+        assert memory is not None, f"Memory type {memory_type} not found"
+        return memory
+
+    async def add(self, memory_type, message, name=None):
+        return await self.get_memory(memory_type=memory_type).add(message, name=name)
+
+    async def modify(self, memory_type, old_content, new_content, name=None):
+        return await self.get_memory(memory_type=memory_type).modify(old_content=old_content, new_content=new_content, name=name)
+
+    async def search(self, memory_type, query, page, start_date=None, end_date=None):
+        return await self.get_memory(memory_type=memory_type).search(query=query, page=page, start_date=start_date, end_date=end_date)

From 395cfe9a09cd135c3a8a08028f4f8d8979a90e56 Mon Sep 17 00:00:00 2001
From: Gong Junmin <1836678486@qq.com>
Date: Thu, 2 Nov 2023 15:16:38 +0800
Subject: [PATCH 56/56] feat: support group chat base

---
 creator/agents/base.py                  |   9 +-
 creator/agents/creator_agent.py         |   5 +-
 creator/agents/extractor_agent.py       |   5 +-
 creator/agents/group_chat.py            | 118 ++++++++++++++++++++++++
 creator/agents/interpreter_agent.py     |   5 +-
 creator/agents/prompt_enhancer_agent.py |   5 +-
 creator/agents/refactor_agent.py        |   5 +-
 creator/agents/tester_agent.py          |   5 +-
 creator/memgpt/agent.py                 |   5 +-
 9 files changed, 132 insertions(+), 30 deletions(-)
 create mode 100644 creator/agents/group_chat.py

diff --git a/creator/agents/base.py b/creator/agents/base.py
index 3a0f186..3985b92 100644
--- a/creator/agents/base.py
+++ b/creator/agents/base.py
@@ -23,13 +23,18 @@ class BaseAgent(LLMChain):
     system_template: str = ""
     allow_user_confirm: bool = False
     prompt: ChatPromptTemplate = ChatPromptTemplate.from_messages(messages=["system", ""])
+    agent_name: str = "BaseAgent"
+    share_memory: bool = False
 
     @property
     def _chain_type(self):
-        return "BaseAgent"
+        return self.agent_name
 
     def __repr__(self) -> str:
-        return self._chain_type + "()"
+        return self.agent_name + "()"
+    
+    def __hash__(self):
+        return hash(self.agent_name)
 
     @property
     def input_keys(self) -> List[str]:
diff --git a/creator/agents/creator_agent.py b/creator/agents/creator_agent.py
index 7791289..445c2ce 100644
--- a/creator/agents/creator_agent.py
+++ b/creator/agents/creator_agent.py
@@ -28,10 +28,7 @@
 class CreatorAgent(BaseAgent):
     total_tries: int = 5
     allow_user_confirm: bool = config.run_human_confirm
-
-    @property
-    def _chain_type(self):
-        return "CreatorAgent"
+    agent_name: str = "CreatorAgent"
 
     def prep_inputs(self, inputs: Dict[str, Any] | Any) -> Dict[str, str]:
         inputs["OPEN_CREATOR_API_DOC"] = OPEN_CREATOR_API_DOC
diff --git a/creator/agents/extractor_agent.py b/creator/agents/extractor_agent.py
index 228a014..4d10f11 100644
--- a/creator/agents/extractor_agent.py
+++ b/creator/agents/extractor_agent.py
@@ -11,10 +11,7 @@
 
 class SkillExtractorAgent(BaseAgent):
     output_key: str = "extracted_skill"
-
-    @property
-    def _chain_type(self):
-        return "SkillExtractorAgent"
+    agent_name: str = "SkillExtractorAgent"
 
     def construct_prompt(self, langchain_messages: Dict[str, Any]):
         prompt = ChatPromptTemplate.from_messages(messages=[
diff --git a/creator/agents/group_chat.py b/creator/agents/group_chat.py
new file mode 100644
index 0000000..52ec00b
--- /dev/null
+++ b/creator/agents/group_chat.py
@@ -0,0 +1,118 @@
+from typing import List, Dict, Any, Optional
+import networkx as nx
+
+from creator.agents.base import BaseAgent
+from creator.utils import print
+
+from langchain.chains.base import Chain
+from langchain.callbacks.manager import CallbackManagerForChainRun
+
+
+class GroupChat(Chain):
+    graph: Optional[nx.DiGraph] = None
+    max_consecutive_auto_reply: int = 3
+
+    @property
+    def input_keys(self) -> List[str]:
+        """Keys expected to be in the chain input."""
+        return ["messages", "sender", "receivers"]
+
+    @property
+    def output_keys(self) -> List[str]:
+        """Keys expected to be in the chain output."""
+        return ["messages", "sender", "receivers"]
+
+    def add_agent(self, agent: BaseAgent):
+        """
+        Add an agent to the graph.
+
+        :param agent: The agent to be added.
+        """
+        self.graph.add_node(agent.agent_name, agent=agent)
+
+    def add_agents(self, agents: List[BaseAgent]):
+        """
+        Add a list of agents to the graph.
+
+        :param agents: The list of agents to be added.
+        """
+        for agent in agents:
+            self.add_agent(agent)
+
+    def remove_agent(self, agent_name: str):
+        """
+        Remove an agent from the graph.
+
+        :param agent_name: The name of the agent to be removed.
+        """
+        if agent_name in self.graph:
+            self.graph.remove_node(agent_name)
+
+    def add_edge(self, agent1: BaseAgent, agent2: BaseAgent):
+        """
+        Add an edge between two agents.
+
+        :param agent1: The first agent.
+        :param agent2: The second agent.
+        """
+        self.graph.add_edge(agent1.agent_name, agent2.agent_name)
+
+    def remove_edge(self, agent1: BaseAgent, agent2: BaseAgent):
+        """
+        Remove an edge between two agents.
+
+        :param agent1: The first agent.
+        :param agent2: The second agent.
+        """
+        self.graph.remove_edge(agent1.agent_name, agent2.agent_name)
+
+    @classmethod
+    def from_mapping(cls, mapping: Dict[BaseAgent, List[BaseAgent]]):
+        graph = nx.DiGraph()
+        node_set = set()
+        for from_node, to_nodes in mapping.items():
+            if from_node.agent_name not in node_set:
+                node_set.add(from_node)
+            for node in to_nodes:
+                if node.agent_name not in node_set:
+                    node_set.add(node)
+                graph.add_edge(from_node.agent_name, node.agent_name)
+        cls.graph = graph
+        return cls
+
+    def _call(
+        self,
+        inputs: Dict[str, Any],
+        run_manager: Optional[CallbackManagerForChainRun] = None,
+    ) -> Dict[str, Any]:
+        return self.run_chat(inputs["messages"], inputs["sender"], inputs["receivers"])
+
+    def run_chat(self, messages: List[Dict], sender: str, receivers: List[str], run_manager: Optional[CallbackManagerForChainRun] = None):
+        """Run a group chat."""
+        assert len(messages) > 0, "Input Messages Must Not Be Empty"
+        curr_cnt = 0
+        while curr_cnt < self.max_consecutive_auto_reply:
+            for receiver in receivers:
+                try:
+                    receiver_agent = self.graph.nodes[receiver]
+                except KeyError:
+                    print("> agent {receiver} not found", print_type="markdown")
+                    raise KeyError
+                if not receiver_agent.share_memory:
+                    messages = [messages[-1]]
+                output_messages = receiver_agent.with_config({"callbacks": run_manager.get_child()}).invoke({"messages": messages})
+                sender = receiver
+                try:
+                    receiver = output_messages[-1]["receiver"]
+                except KeyError:
+                    print("> agent {receiver} has no receiver", print_type="markdown")
+                    raise KeyError
+
+                if not receiver_agent.share_memory:
+                    messages = messages[:-1] + output_messages
+                else:
+                    messages = output_messages
+
+                if receiver == "human":
+                    return messages, sender, receiver
+                curr_cnt += 1
diff --git a/creator/agents/interpreter_agent.py b/creator/agents/interpreter_agent.py
index c03c7c7..e0a31b3 100644
--- a/creator/agents/interpreter_agent.py
+++ b/creator/agents/interpreter_agent.py
@@ -18,11 +18,8 @@
 class CodeInterpreterAgent(BaseAgent):
     total_tries: int = 10
     allow_user_confirm: bool = config.run_human_confirm
+    agent_name: str = "CodeInterpreterAgent"
 
-    @property
-    def _chain_type(self):
-        return "CodeInterpreterAgent"
-    
     def postprocess_mesasge(self, message):
         function_call = message.additional_kwargs.get("function_call", None)
         if function_call is not None:
diff --git a/creator/agents/prompt_enhancer_agent.py b/creator/agents/prompt_enhancer_agent.py
index 4faa7fc..703a6fd 100644
--- a/creator/agents/prompt_enhancer_agent.py
+++ b/creator/agents/prompt_enhancer_agent.py
@@ -12,10 +12,7 @@
 
 class PromptEnhancerAgent(BaseAgent):
     output_key: str = "request"
-
-    @property
-    def _chain_type(self):
-        return "PromptEnhancerAgent"
+    agent_name: str = "PromptEnhancerAgent"
 
     def construct_prompt(self, langchain_messages: Dict[str, Any]):
         prompt = ChatPromptTemplate.from_messages(messages=[
diff --git a/creator/agents/refactor_agent.py b/creator/agents/refactor_agent.py
index cdcb2b2..49176ad 100644
--- a/creator/agents/refactor_agent.py
+++ b/creator/agents/refactor_agent.py
@@ -12,10 +12,7 @@
 
 class CodeRefactorAgent(BaseAgent):
     output_key: str = "refacted_skills"
-
-    @property
-    def _chain_type(self):
-        return "CodeRefactorAgent"
+    agent_name: str = "CodeRefactorAgent"
 
     def construct_prompt(self, langchain_messages: Dict[str, Any]):
         prompt = ChatPromptTemplate.from_messages(messages=[
diff --git a/creator/agents/tester_agent.py b/creator/agents/tester_agent.py
index f80b2ff..c1754b1 100644
--- a/creator/agents/tester_agent.py
+++ b/creator/agents/tester_agent.py
@@ -19,10 +19,7 @@ class CodeTesterAgent(BaseAgent):
     total_tries: int = 10
     output_key: str = "output"
     allow_user_confirm: bool = config.run_human_confirm
-
-    @property
-    def _chain_type(self):
-        return "CodeTesterAgent"
+    agent_name: str = "CodeTesterAgent"
 
     def postprocess_mesasge(self, message):
         function_call = message.additional_kwargs.get("function_call", None)
diff --git a/creator/memgpt/agent.py b/creator/memgpt/agent.py
index 5d4468f..5e9badc 100644
--- a/creator/memgpt/agent.py
+++ b/creator/memgpt/agent.py
@@ -36,10 +36,7 @@ class MemGPT(BaseAgent):
     memgpt_config: dict = None
     pause_heartbeats_start: datetime.datetime = None
     pause_heartbeats_minutes: int = 0
-
-    @property
-    def _chain_type(self):
-        return "MemGPT"
+    agent_name: str = "MemGPT"
 
     @property
     def input_keys(self) -> List[str]: