Merge pull request #92 from fractalego/prompt-as-list

Prompt as list

documentation/source/configuration.rst

@@ -13,9 +13,11 @@ A typical configuration file looks like this:
       "deactivate_sound": true,
       "rules": "rules.yaml",
       "functions": "functions.py",
+      "frontend_port": 8081,
       "llm_model": {
         "model_host": "localhost",
-        "model_port": 8080
+        "model_port": 8080,
+        "temperature": 0.4
       },
       "listener_model": {
         "model_host": "localhost",
@@ -45,7 +47,10 @@ These settings regulate the following:
 
     * "functions" is the file containing the functions that can be used in the rules. The default is "functions.py".
 
+    * "frontend_port" is the port where the web frontend is running. The default is 8090.
+
     * "llm_model" is the configuration to connect to the LLM model in the backend. The default is "localhost:8080".
+       The "temperature" parameter is used to set the temperature for the LLM model. The default is 0.4.
 
     * "listener_model" is the configuration to connect to the listener model in the backend. The default is "localhost:8080".
 

tests/test_rules.py

@@ -88,9 +88,8 @@ def test__nested_rules_are_printed_correctly(self):
         rule = facts_and_rules["rules"][0]
         expected = """
 the user wants to know the time
-  -output "The time is <execute>get_time()</execute>"
-    -if the time is before 12:00 say "Good morning"
-    -if the time is after 12:00 say "Good afternoon"        
+  - output "The time is <execute>get_time()</execute>"
+    - if the time is before 12:00 say "Good morning"
+    - if the time is after 12:00 say "Good afternoon"        
         """.strip()
         self.assertEqual(expected, str(rule).strip())
-

todo.txt

@@ -1,3 +1,17 @@
+* add config file for model names
+    - llm model name
+    - whisper model name
+
+/* add version name
+/* let user decide port for frontend
+/* update docs about port
+/* push new version
+* update pypi with wafl and wafl-llm
+* clean code for llm eval and make it public
+* update huggingface readme
+* read overleaf paper
+
+
 * on wafl_llm make it so only some LLMs are supported
 * change speaker model with newer one
 

wafl/answerer/dialogue_answerer.py

@@ -160,9 +160,19 @@ def _init_python_module(self, module_name):
         self._functions = [item[0] for item in getmembers(self._module, isfunction)]
 
     async def _substitute_results_in_answer(self, answer_text):
-        matches = re.finditer(r"<execute>(.*?)</execute>", answer_text, re.DOTALL)
+        matches = re.finditer(r"<execute>(.*?)</execute>|<execute>(.*?\))$", answer_text, re.DOTALL|re.MULTILINE)
         for match in matches:
             to_execute = match.group(1)
+            if not to_execute:
+                continue
+            result = await self._run_code(to_execute)
+            answer_text = answer_text.replace(match.group(0), result)
+
+        matches = re.finditer(r"<execute>(.*?\))$", answer_text, re.DOTALL|re.MULTILINE)
+        for match in matches:
+            to_execute = match.group(1)
+            if not to_execute:
+                continue
             result = await self._run_code(to_execute)
             answer_text = answer_text.replace(match.group(0), result)
 
@@ -178,6 +188,13 @@ async def _substitute_memory_in_answer_and_get_memories_if_present(
             answer_text = answer_text.replace(match.group(0), "")
             memories.append(to_execute)
 
+        matches = re.finditer(r"<remember>(.*?\))$", answer_text, re.DOTALL|re.MULTILINE)
+        memories = []
+        for match in matches:
+            to_execute = match.group(1)
+            answer_text = answer_text.replace(match.group(0), "")
+            memories.append(to_execute)
+
         return answer_text, memories
 
     async def _run_code(self, to_execute):
@@ -206,7 +223,7 @@ async def _run_code(self, to_execute):
 
             except Exception as e:
                 result = (
-                    f'Error while executing\n\n```python\n{to_execute}\n```\n\n{str(e)}'
+                    f"Error while executing\n\n```python\n{to_execute}\n```\n\n{str(e)}"
                 )
                 traceback.print_exc()
                 break

wafl/answerer/rule_maker.py

@@ -1,12 +1,12 @@
 class RuleMaker:
     def __init__(
         self,
-        knowledge,
-        config,
-        interface,
-        max_num_rules,
-        delete_current_rule,
-        max_recursion=3,
+        knowledge: "Knowledge",
+        config: "BaseConfig",
+        interface: "BaseInterface",
+        max_num_rules: int,
+        delete_current_rule: str,
+        max_recursion: int = 3,
     ):
         self._knowledge = knowledge
         self._config = config
@@ -26,6 +26,8 @@ async def create_from_query(self, query):
             rules_text = rule.get_string_using_template(
                 "- If {effect} go through the following points:"
             )
+            rules_text += f'{rule.indent_str}- After you completed all the steps output "{self._delete_current_rule}" and continue the conversation.\n'
+
             rules_texts.append(rules_text)
             await self._interface.add_fact(f"The bot remembers the rule:\n{rules_text}")
 

wafl/connectors/base_llm_connector.py

@@ -1,7 +1,6 @@
 import logging
 import re
 
-
 from wafl.connectors.utils import select_best_answer
 
 _system_logger = logging.getLogger(__file__)
@@ -39,11 +38,7 @@ async def generate(self, prompt: str) -> str:
 
         text = prompt
         start = len(text)
-        while (
-            all(item not in text[start:] for item in self._last_strings)
-            and len(text) < start + self._max_reply_length
-        ):
-            text += select_best_answer(await self.predict(text), self._last_strings)
+        text += select_best_answer(await self.predict(text), self._last_strings)
 
         end_set = set()
         for item in self._last_strings:

wafl/connectors/remote/remote_llm_connector.py

@@ -1,7 +1,10 @@
+import json
+
 import aiohttp
 import asyncio
 
 from wafl.connectors.base_llm_connector import BaseLLMConnector
+from wafl.variables import is_supported
 
 
 class RemoteLLMConnector(BaseLLMConnector):
@@ -14,6 +17,7 @@ def __init__(self, config, last_strings=None, num_replicas=3):
         super().__init__(last_strings)
         host = config["model_host"]
         port = config["model_port"]
+        self._default_temperature = config["temperature"]
         self._server_url = f"https://{host}:{port}/predictions/bot"
         self._num_replicas = num_replicas
 
@@ -28,9 +32,11 @@ def __init__(self, config, last_strings=None, num_replicas=3):
         ):
             raise RuntimeError("Cannot connect a running LLM.")
 
-    async def predict(self, prompt: str, temperature=None, num_tokens=None, num_replicas=None) -> [str]:
+    async def predict(
+        self, prompt: str, temperature=None, num_tokens=None, num_replicas=None
+    ) -> [str]:
         if not temperature:
-            temperature = 0.5
+            temperature = self._default_temperature
 
         if not num_tokens:
             num_tokens = self._num_prediction_tokens
@@ -52,8 +58,11 @@ async def predict(self, prompt: str, temperature=None, num_tokens=None, num_repl
                 connector=aiohttp.TCPConnector(ssl=False),
             ) as session:
                 async with session.post(self._server_url, json=payload) as response:
-                    answer = await response.text()
-                    return answer.split("<||>")
+                    answer = json.loads(await response.text())
+                    status = answer["status"]
+                    if status != "success":
+                        raise RuntimeError(f"Error in prediction: {answer}")
+                    return answer["prediction"].split("<||>")
 
         return [""]
 
@@ -70,7 +79,14 @@ async def check_connection(self):
                 conn_timeout=3, connector=aiohttp.TCPConnector(ssl=False)
             ) as session:
                 async with session.post(self._server_url, json=payload) as response:
-                    await response.text()
+                    answer = json.loads(await response.text())
+                    wafl_llm_version = answer["version"]
+                    print(f"Connected to wafl-llm v{wafl_llm_version}.")
+                    if not is_supported(wafl_llm_version):
+                        print("This version of wafl-llm is not supported.")
+                        print("Please update wafl-llm.")
+                        raise aiohttp.client.InvalidURL
+
                     return True
 
         except aiohttp.client.InvalidURL:

wafl/connectors/remote/remote_whisper_connector.py

@@ -39,7 +39,9 @@ async def predict(self, waveform, hotword=None) -> Dict[str, float]:
                     data = await response.text()
                     prediction = json.loads(data)
                     if "transcription" not in prediction:
-                        raise RuntimeError("No transcription found in prediction. Is your microphone working?")
+                        raise RuntimeError(
+                            "No transcription found in prediction. Is your microphone working?"
+                        )
 
                     transcription = prediction["transcription"]
                     score = prediction["score"]

wafl/interface/base_interface.py

@@ -70,5 +70,7 @@ def _decorate_reply(self, text: str) -> str:
         return self._decorator.extract(text, self._utterances)
 
     def _insert_utterance(self, speaker, text: str):
-        if self._utterances == [] or text != self._utterances[-1][1].replace(f"{speaker}: ", ""):
+        if self._utterances == [] or text != self._utterances[-1][1].replace(
+            f"{speaker}: ", ""
+        ):
             self._utterances.append((time.time(), f"{speaker}: {text}"))

wafl/interface/list_interface.py

@@ -13,20 +13,20 @@ def __init__(self, interfaces_list: List[BaseInterface]):
     async def output(self, text: str, silent: bool = False):
         await asyncio.wait(
             [interface.output(text, silent) for interface in self._interfaces_list],
-            return_when=asyncio.ALL_COMPLETED
+            return_when=asyncio.ALL_COMPLETED,
         )
 
     async def input(self) -> str:
         done, pending = await asyncio.wait(
             [interface.input() for interface in self._interfaces_list],
-            return_when=asyncio.FIRST_COMPLETED
+            return_when=asyncio.FIRST_COMPLETED,
         )
         return done.pop().result()
 
     async def insert_input(self, text: str):
         await asyncio.wait(
             [interface.insert_input(text) for interface in self._interfaces_list],
-            return_when=asyncio.ALL_COMPLETED
+            return_when=asyncio.ALL_COMPLETED,
         )
 
     def bot_has_spoken(self, to_set: bool = None):
@@ -44,11 +44,10 @@ def deactivate(self):
         super().deactivate()
         self._synchronize_interfaces()
 
-
     def add_hotwords(self, hotwords):
         for interface in self._interfaces_list:
             interface.add_hotwords(hotwords)
 
     def _synchronize_interfaces(self):
         for interface in self._interfaces_list:
-            interface._utterances = self._utterances
+            interface._utterances = self._utterances

wafl/rules.py

@@ -7,7 +7,7 @@ class Rule:
     effect: "Fact"
     causes: List["Fact"]
     _max_indentation = 3
-    _indent_str = "  "
+    indent_str = "  "
 
     def toJSON(self):
         return str(self)
@@ -33,22 +33,21 @@ def _add_clauses(self, rule_str: str) -> str:
 
         return rule_str
 
-
-    def _recursively_add_clauses(self, query: str, depth: int=1) -> str:
-        indentation = self._indent_str * depth
+    def _recursively_add_clauses(self, query: str, depth: int = 1) -> str:
+        indentation = self.indent_str * depth
         if type(query) == str:
-            return f"{indentation}-{query}\n"
+            return f"{indentation}- {query}\n"
 
         if type(query.text) == str:
-            return f"{indentation}-{query.text}\n"
+            return f"{indentation}- {query.text}\n"
 
         if depth > self._max_indentation:
             return ""
 
         clause = list(query.text.keys())[0]
-        rules_text = f"{indentation}-{clause}\n"
+        rules_text = f"{indentation}- {clause}\n"
         for clauses in query.text.values():
             for cause_index, clause in enumerate(clauses):
                 rules_text += self._recursively_add_clauses(clause, depth + 1)
 
-        return rules_text
+        return rules_text

wafl/runners/routes.py

@@ -77,4 +77,4 @@ def add_new_rules(app: Flask, conversation_id: int, web_server_loop: "WebLoop"):
         f"toggle_logs_{conversation_id}",
         web_server_loop.toggle_logs,
         methods=["POST"],
-    )
+    )

wafl/runners/run_web_and_audio_interface.py

@@ -1,4 +1,3 @@
-import asyncio
 import random
 import sys
 import threading
@@ -56,7 +55,7 @@ def create_scheduler_and_webserver_loop(conversation_id):
             "web_server_loop": web_loop,
         }
 
-    app.run(host="0.0.0.0", port=8889)
+    app.run(host="0.0.0.0", port=Configuration.load_local_config().get_value("frontend_port"))
 
 
 if __name__ == "__main__":

wafl/runners/run_web_interface.py

@@ -57,7 +57,7 @@ def create_scheduler_and_webserver_loop(conversation_id):
             "web_server_loop": web_loop,
         }
 
-    app.run(host="0.0.0.0", port=8889)
+    app.run(host="0.0.0.0", port=Configuration.load_local_config().get_value("frontend_port"))
 
 
 if __name__ == "__main__":

wafl/templates/config.json

@@ -4,9 +4,11 @@
   "deactivate_sound": true,
   "rules": "rules.yaml",
   "functions": "functions.py",
+  "frontend_port": 8090,
   "llm_model": {
     "model_host": "localhost",
-    "model_port": 8080
+    "model_port": 8080,
+    "temperature": 0.4
   },
   "listener_model": {
     "model_host": "localhost",

wafl/templates/rules.yaml

@@ -45,4 +45,12 @@ rules:
 
   - the user asks something about cities, capitals, countries, buildings, famous people, bars, restaurants, rivers, mountains, lakes, seas, oceans, planets, stars, galaxies:
       - say that you are just improvising the answer
-      - say what you think answer the question
+      - say what you think answer the question
+
+  - the user thanks the bot:
+    - The intention of the user is to close the conversation
+    - You must answer the user by writing "<execute>close_conversation()</execute>"
+
+  - the user says "end this conversation":
+    - The intention of the user is to close the conversation
+    - You must answer the user by writing "<execute>close_conversation()</execute>"

wafl/variables.py

@@ -1,4 +1,9 @@
 def get_variables():
     return {
-        "version": "0.0.82",
+        "version": "0.0.83",
     }
+
+
+def is_supported(wafl_llm_version):
+    supported_versions = ["0.0.82", "0.0.83"]
+    return wafl_llm_version in supported_versions