added prompt file + docs

documentation/source/configuration.rst

@@ -12,8 +12,11 @@ A typical configuration file looks like this:
       "waking_up_sound": true,
       "deactivate_sound": true,
       "rules": "rules.yaml",
+      "index": "indices.yaml",
+      "cache_filename": "knowledge_cache",
+      "prompt_filename": "main.prompt",
       "functions": "functions.py",
-      "frontend_port": 8081,
+      "max_recursion": 2,
       "llm_model": {
         "model_host": "localhost",
         "model_port": 8080,
@@ -37,6 +40,7 @@ A typical configuration file looks like this:
     }
 
 
+
 These settings regulate the following:
 
     * "waking_up_word" is the name of the bot, used to wake up the system in the "run-audio" mode.
@@ -45,6 +49,12 @@ These settings regulate the following:
 
     * "rules" is the file containing the facts and rules that guide the chatbot. The default is "rules.yaml".
 
+    * "index" is the file containing the path to the files to index. The default is "indices.yaml".
+
+    * "cache_filename" is the file where the indexed knowledge is cached. The default is "knowledge_cache".
+
+    * "prompt_filename" is the file containing the main prompt for the chatbot. The default is "main.prompt".
+
     * "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.

documentation/source/index.rst

@@ -16,6 +16,7 @@ Welcome to WAFL's 0.0.90 documentation!
    configuration
    running_WAFL
    facts_and_rules
+   modify_the_prompt
    examples
    testcases
    actions

documentation/source/modify_the_prompt.rst

@@ -0,0 +1,23 @@
+Modify the original prompt
+==========================
+
+The prompt is stored in the file "main.prompt" in the project's root directory.
+The name of the file can be changed in the `config.json` file.
+The default is:
+
+
+.. code-block:: text
+    A user is chatting with a bot. The chat is happening through a web interface. The user is typing the messages and the bot is replying.
+
+    This is summary of the bot's knowledge:
+    {facts}
+
+    The rules that *must* be followed are:
+    {rules}
+
+    Create a plausible dialogue based on the aforementioned summary and rules.
+    Do not repeat yourself. Be friendly but not too servile.
+    Follow the rules if present and they apply to the dialogue. Do not improvise if rules are present.
+
+
+The variables `{facts}` and `{rules}` are replaced by the actual facts and rules when the prompt is generated.

tests/config.json

@@ -4,7 +4,8 @@
   "deactivate_sound": true,
   "rules": "rules.yaml",
   "index": "indices.yaml",
-  "index_filename": "knowledge_cache",
+  "cache_filename": "knowledge_cache",
+  "prompt_filename": "main.prompt",
   "functions": "functions.py",
   "max_recursion": 2,
   "llm_model": {

tests/main.prompt

@@ -0,0 +1,11 @@
+A user is chatting with a bot. The chat is happening through a web interface. The user is typing the messages and the bot is replying.
+
+This is summary of the bot's knowledge:
+{facts}
+
+The rules that *must* be followed are:
+{rules}
+
+Create a plausible dialogue based on the aforementioned summary and rules.
+Do not repeat yourself. Be friendly but not too servile.
+Follow the rules if present and they apply to the dialogue. Do not improvise if rules are present.

todo.txt

@@ -1,12 +1,21 @@
+* interruptible speech
+* dependabot!!!
+* use poetry
+
+PharazonE
+* upload to hetzner and make it work for some retrieval tasks
+* develop more rules + use-cases for voice and other
+
+
 /* add control over which llm to use from the frontend
 /  - add list of models in the backend
 
-* interruptible speech
-* add option so use llama.cpp from wafl_llm
-* add option to have None as a model setting in wafl_llm
+/* add quantization of llm to wafl_llm config
+/* write docs about it on wafl
+
+/* add option so use llama.cpp from wafl_llm
+/* add option to have None as a model setting in wafl_llm
 
-* dependabot!!!
-* use poetry
 /* add pdf to indexing
 * add json to indexing
 /* add metadata to indexing items

wafl/answerer/dialogue_answerer.py

@@ -11,7 +11,7 @@
 )
 from wafl.answerer.base_answerer import BaseAnswerer
 from wafl.answerer.rule_maker import RuleMaker
-from wafl.connectors.clients.llm_chitchat_answer_client import LLMChitChatAnswerClient
+from wafl.connectors.clients.llm_chat_client import LLMChatClient
 from wafl.dataclasses.dataclasses import Query, Answer
 from wafl.interface.conversation import Conversation
 from wafl.simple_text_processing.questions import is_question
@@ -20,7 +20,7 @@
 class DialogueAnswerer(BaseAnswerer):
     def __init__(self, config, knowledge, interface, code_path, logger):
         self._threshold_for_facts = 0.85
-        self._client = LLMChitChatAnswerClient(config)
+        self._client = LLMChatClient(config)
         self._knowledge = knowledge
         self._logger = logger
         self._interface = interface

wafl/connectors/clients/llm_chitchat_answer_client.py → wafl/connectors/clients/llm_chat_client.py

@@ -1,5 +1,4 @@
 import os
-import textwrap
 from typing import List
 
 from wafl.connectors.factories.llm_connector_factory import LLMConnectorFactory
@@ -8,10 +7,12 @@
 _path = os.path.dirname(__file__)
 
 
-class LLMChitChatAnswerClient:
+class LLMChatClient:
     def __init__(self, config):
         self._connector = LLMConnectorFactory.get_connector(config)
         self._config = config
+        with open(self._config.get_value("prompt_filename")) as f:
+            self.prompt = f.read()
 
     async def get_answer(self, text: str, dialogue: str, rules_text: List[str]) -> str:
         prompt = await self._get_answer_prompt(text, dialogue, "\n".join(rules_text))
@@ -26,16 +27,4 @@ async def _get_answer_prompt(
         )
 
     def _get_system_prompt(self, text, rules_text):
-        return f"""
-A user is chatting with a bot. The chat is happening through a web interface. The user is typing the messages and the bot is replying.        
-
-This is summary of the bot's knowledge: 
-{text.strip()}
-
-The rules that *must* be followed are:
-{rules_text.strip()}
-
-Create a plausible dialogue based on the aforementioned summary and rules. 
-Do not repeat yourself. Be friendly but not too servile.
-Follow the rules if present and they apply to the dialogue. Do not improvise if rules are present. 
-        """.strip()
+        return self.prompt.replace("{facts}", text.strip()).replace("{rules}", rules_text.strip()).strip()

wafl/knowledge/indexing_implementation.py

@@ -36,14 +36,14 @@ async def load_knowledge(config, logger=None):
     with open(index_filename) as file:
         index_txt = file.read()
 
-    if os.path.exists(config.get_value("index_filename")):
-        knowledge = joblib.load(config.get_value("index_filename"))
+    if os.path.exists(config.get_value("cache_filename")):
+        knowledge = joblib.load(config.get_value("cache_filename"))
         if knowledge.hash == hash(rules_txt + index_txt):
             return knowledge
 
     knowledge = SingleFileKnowledge(config, rules_txt, logger=logger)
     knowledge = await _add_indices_to_knowledge(knowledge, index_txt)
-    joblib.dump(knowledge, config.get_value("index_filename"))
+    joblib.dump(knowledge, config.get_value("cache_filename"))
     await knowledge.initialize_retrievers()
     return knowledge
 

wafl/templates/config.json

@@ -4,7 +4,8 @@
   "deactivate_sound": true,
   "rules": "rules.yaml",
   "index": "indices.yaml",
-  "index_filename": "knowledge_cache",
+  "cache_filename": "knowledge_cache",
+  "prompt_filename": "main.prompt",
   "functions": "functions.py",
   "max_recursion": 2,
   "frontend_port": 8090,

wafl/templates/main.prompt

@@ -0,0 +1,11 @@
+A user is chatting with a bot. The chat is happening through a web interface. The user is typing the messages and the bot is replying.
+
+This is summary of the bot's knowledge:
+{facts}
+
+The rules that *must* be followed are:
+{rules}
+
+Create a plausible dialogue based on the aforementioned summary and rules.
+Do not repeat yourself. Be friendly but not too servile.
+Follow the rules if present and they apply to the dialogue. Do not improvise if rules are present.