Refactored the retrieval chain, and MiniPilot to Minipilot

mortensi · mortensi · commit 8a7391f3e2d0 · 2024-10-16T13:53:18.000+02:00
diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-# MiniPilot
+# Minipilot
 
 This application implements a chatbot you can train with your data. From the GUI you will be able to:
 
@@ -16,7 +16,7 @@ The system uses:
 
 ## Quickstart
 
-Follow these instructions to build and start MiniPilot and a Redis Stack database as containers on your laptop. For more installation methods, refer to the docs.
+Follow these instructions to build and start Minipilot and a Redis Stack database as containers on your laptop. For more installation methods, refer to the docs.
 
 1. `git clone https://github.com/redis/minipilot.git`
 2. `export OPENAI_API_KEY="your-openai-key"`
diff --git a/docs/architecture.md b/docs/architecture.md
@@ -15,6 +15,6 @@ The explanation of the steps in the pipeline follows.
 7. The question and answer pair is added to the cache if relevant (a basic criterion is represented by the context. If there was at least a result while fetching the context for RAG, then the question is pertinent, and the answer can be saved).
 8. The interaction is added to the conversation history.
 
-Regarding the interactions with OpenAI, MiniPilot uses the [chat completion api](https://platform.openai.com/docs/guides/chat-completions) via the LangChain [ChatOpenAI](https://python.langchain.com/v0.2/docs/integrations/chat/openai/) class as well as the `text-embedding-ada-002` embedding model for the retrievers and the semantic cache.
+Regarding the interactions with OpenAI, Minipilot uses the [chat completion api](https://platform.openai.com/docs/guides/chat-completions) via the LangChain [ChatOpenAI](https://python.langchain.com/v0.2/docs/integrations/chat/openai/) class as well as the `text-embedding-ada-002` embedding model for the retrievers and the semantic cache.
 
 ![architecture](../src/static/images/minipilot-ai-pipeline.png)
diff --git a/docs/faq.md b/docs/faq.md
@@ -2,21 +2,21 @@
 
 ## What is this?
 
-MiniPilot is a GenAI-assisted chatbot. This demo is a minimal implementation (hence the name "MiniPilot"!) which allows you to import data as CSV, process it and store it in a Redis vector database.
-You can clone and test the repository, fork it, change the code and, if you like, contribute to the repository. It can be extended to import data from other document formats (like PDF) or crawl and scrape web resources. Ad-hoc connectors can be developed to extend the MiniPilot.
+Minipilot is a GenAI-assisted chatbot. This demo is a minimal implementation (hence the name "Minipilot"!) which allows you to import data as CSV, process it and store it in a Redis vector database.
+You can clone and test the repository, fork it, change the code and, if you like, contribute to the repository. It can be extended to import data from other document formats (like PDF) or crawl and scrape web resources. Ad-hoc connectors can be developed to extend the Minipilot.
 
-## What features can I find in MiniPilot?
+## What features can I find in Minipilot?
 
-MiniPilot supports:
+Minipilot supports:
 
 - Conversation history by session, to make sense of the interactions of a user
 - Semantic cache, to store and retrieve past interactions (from all the conversations of all the users). This speeds up the system, because if an answer is in the cache, there is no need to invoke the OpenAI chat endpoint at all.
 - Rate limiter, a basic implementation based on Redis
 - Configurability, to tune up every single piece of the chatbot
-- REST API: the MiniPilot can serve remote users over a REST API. The UI is for demonstrative purposes. 
+- REST API: the Minipilot can serve remote users over a REST API. The UI is for demonstrative purposes. 
 - Centralized Redis session, so application servers are purely stateless, which helps scalability of the service
 
-## What can I do with MiniPilot?
+## What can I do with Minipilot?
 
 You can use the code for self-learning, extend it as desired and use in production. The source code sheds light on all the standard components used in these kinds of systems:
 
diff --git a/docs/installation.md b/docs/installation.md
@@ -15,7 +15,7 @@ Then, decide where you want to run your Redis Stack instance.
 
 ## Local setup
 
-To run MiniPilot on your laptop, just clone the repository.
+To run Minipilot on your laptop, just clone the repository.
 
 ```commandline
 git clone https://github.com/redis/MiniPilot.git
@@ -81,10 +81,10 @@ DB_PORT                The port of your Redis Stack instance or Redis Cloud data
 DB_PWD                 The password of your Redis Stack instance or Redis Cloud database
 DB_SSL                 If you would like to connect using SSL (mandatory for Redis Cloud databases)
 OPENAI_API_KEY         Your OpenAI token
-MINIPILOT_ENDPOINT     MiniPilot is a REST API service, but the internal chat interface uses the REST API, too, on the default exposed port 8000. You can just leave http://127.0.0.1:8000
+MINIPILOT_ENDPOINT     Minipilot is a REST API service, but the internal chat interface uses the REST API, too, on the default exposed port 8000. You can just leave http://127.0.0.1:8000
 ```
 
-## Configuring MiniPilot
+## Configuring Minipilot
 
 1. point your browser to [http://127.0.0.1:5005/](http://127.0.0.1:5005/)
 2. From the `data` tab, load a CSV dataset, and index it. You can try the [IMDB movies dataset](https://www.kaggle.com/datasets/ashpalsingh1525/imdb-movies-dataset)
diff --git a/docs/usage.md b/docs/usage.md
@@ -1,6 +1,6 @@
 # Usage
 
-MiniPilot is a highly configurable server. Read the environment variables to customize the server's behavior.
+Minipilot is a highly configurable server. Read the environment variables to customize the server's behavior.
 
 | Variable                               | Description                                                                                                                    | Default           |
 |----------------------------------------|--------------------------------------------------------------------------------------------------------------------------------|-------------------|
diff --git a/src/core/RedisRetrievalChain.py b/src/core/RedisRetrievalChain.py
@@ -92,15 +92,14 @@ def __ask_question(self, q, callback_fn: StreamingStdOutCallbackHandlerYield):
                 callback_fn.q.put(cached[0]['response'])
                 callback_fn.q.put(STOP_ITEM)
 
-                # The question is in the cache, but I want to save the conversation in the history too
+                # The question is in the cache, but we want to save the conversation in the history too
                 metadata = {}
                 if 'metadata' in cached[0]:
                     metadata = cached[0]['metadata']
                 redis_history.add_user_message(q)
                 redis_history.add_message(BaseMessage(content=cached[0]['response'], type="ai", additional_kwargs=metadata))
                 return
 
-        # Note that a try/catch is not needed here. Callback takes care of all errors in `on_llm_error`
         # llm = OpenAI(streaming=True, callbacks=[callback_fn])
         streaming_llm = ChatOpenAI(
             model_name=self.model,
@@ -138,14 +137,14 @@ def get_chat_history(inputs) -> str:
         # now you can rebuild this using https://js.langchain.com/v0.1/docs/use_cases/code_understanding/#chat
         # https://python.langchain.com/v0.2/docs/versions/migrating_chains/conversation_retrieval_chain/
         chatbot = ConversationalRetrievalChain.from_llm(llm=streaming_llm,
-                                                        retriever=self.__get_retriever_with_score(MINIPILOT_CONTEXT_LENGTH),
-                                                        get_chat_history=get_chat_history,
+                                                        retriever=self.__get_retriever_with_score(MINIPILOT_CONTEXT_LENGTH),    # RAG
+                                                        get_chat_history=get_chat_history,                                      # LLM Memory
                                                         rephrase_question=False,
                                                         return_generated_question=True,
                                                         verbose=MINIPILOT_DEBUG,
                                                         return_source_documents=True,
-                                                        condense_question_llm = ChatOpenAI(temperature=0, model=self.model),
-                                                        combine_docs_chain_kwargs={'prompt': qa_prompt})
+                                                        condense_question_llm = ChatOpenAI(temperature=0, model=self.model),    # LLM Memory part 2: condense
+                                                        combine_docs_chain_kwargs={'prompt': qa_prompt})                        # Agentic Memory
 
         result = None
         try:
@@ -162,9 +161,8 @@ def get_chat_history(inputs) -> str:
 
             """
             Update the cache only if caching is enabled AND there was context retrieved for RAG
-            in such a case, we may save the generated standalone question, which also includes summarization of the history
-            but for now we save the original question, or a user repeating over and over a question, will never see his own cached question. 
-            Having references means that there is semantic content available in the database, so the question should make sense
+            First, we save the generated standalone question, which also includes summarization of the history
+            Second, having references means that there is semantic content available in the database, so the question should make sense
             and it can be stored, otherwise the following might happen:
             1. ask a question "can this or that happen?"
             2. get references for RAG, forward the prompt to the LMM and get an answer, hence cache the pair question/asnwer
diff --git a/src/static/.DS_Store b/src/static/.DS_Store
diff --git a/src/templates/botbar.html b/src/templates/botbar.html
@@ -1,7 +1,7 @@
 <form class="search-form mt-4 p-2 pl-4" method="post" action="">
     <div class="field has-addons">
         <div class="control is-expanded is-flex is-vcentered">
-            <input name="q" autocomplete="off" class="search-bar" type="text" placeholder="Ask the MiniPilot a question...">
+            <input name="q" autocomplete="off" class="search-bar" type="text" placeholder="Ask Minipilot a question...">
         </div>
         <div class="control has-icons-left">
             <button id="chat" disabled type="submit" class="button search-button" autocomplete="off">
diff --git a/src/templates/chat.html b/src/templates/chat.html
@@ -8,7 +8,7 @@
         <div class="column is-two-thirds is-full-mobile">
                 <div class="bubble bubble-initial text-lg tracking-tight">
                     <p>
-                        Hello! I am the MiniPilot, how can I help you?
+                        Hello! I am the Minipilot, how can I help you?
                     </p>
                 </div>
             <span id="conversation">
