Create a chat app like chatgpt to allow search in the database v1

[billmetangmo]

Multiples ways of solving:

• Use https://github.com/jerryjliu/gpt_index & https://github.com/hwchase17/langchain ( example https://github.com/arc53/DocsGPT ) . Pour le chargement utilisez un loader https://llamahub.ai/l/file-simple_csv ?
• Supabase Clippy: ChatGPT for Supabase Docs: https://supabase.com/blog/chatgpt-supabase-docs
• Use aista https://aista.hashnode.dev/open-source-chatgpt-based-bing-search-launched
• Use haystack https://github.com/deepset-ai/haystack
• Use Hugging face https://github.com/huggingface/transformers ?
• API de chatgpt ( released on March 2th https://platform.openai.com/docs/guides/chat )
• The smallest possible LLM API : https://github.com/tomdyson/microllama
• OpenAI ChatGPT plugin: https://github.com/openai/chatgpt-retrieval-plugin
• Llama : https://github.com/ggerganov/llama.cpp

[billmetangmo]

Create a simple data app for the first version using streamlit

[billmetangmo]

Python libraries for action-oriented LLMs: https://www.youtube.com/watch?v=XgD5AuzBvs8

[billmetangmo]

Search using the relevant documents provided by semantic search: may be we can use embedding then search using OpenAI ? https://learnprompting.org/docs/applied_prompting/build_chatbot_from_kb ?

[billmetangmo]

Vue des tâches: https://www.loom.com/share/0f04f92d3c254fa4bcfc367121982c8a

• [x] Save faiss embedding to avoid to calculate it at ech execution of ui.py
• [x] Change the prompt so that when a person asks for number of assos , it said to go to tchoung-te.mongulu.cm
• [x] Fine tune the prompt if necessary
• [x] See if possible to have sources in chainlit using RetrievalQAWithSourcesChain from langchain
• [x] Ajouter une image sur chainlit.md si possible
• [x] Déploiement de l'application sur nos serveurs ( @billmetangmo )
• [x] Add langsmith,helicone

[KameniAlexNea]

Here's a formatted explanation of the traceback for other developers:

Traceback (most recent call last):
  File "/home/eak/Documents/AI/DeepLearning/tchoung-te/.venv/bin/chainlit", line 8, in <module>
    sys.exit(cli())
  ...
  ...
  File "pydantic/main.py", line 341, in pydantic.main.BaseModel.__init__
pydantic.error_wrappers.ValidationError: 1 validation error for StuffDocumentsChain
__root__
  document_variable_name summaries was not found in llm_chain input_variables: ['context', 'question'] (type=value_error)

This is the part of the code that failed:

llm = ChatOpenAI(max_tokens=500, temperature=0, model_name="gpt-3.5-turbo")
chain_type_kwargs = {"prompt": CHAT_PROMPT}
qa = RetrievalQAWithSourcesChain.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=vectors.as_retriever(search_kwargs={"k": 3}),
    chain_type_kwargs=chain_type_kwargs,
    return_source_documents=True
)

The traceback indicates an error in the code, specifically in the part where you're initializing the qa object. The error message (pydantic.error_wrappers.ValidationError) suggests that there's a validation error related to the StuffDocumentsChain and its input variables. The error message details that the variable name summaries was not found in the llm_chain input_variables. You might need to check the input variables and ensure that summaries is properly defined or used in your code.

[KameniAlexNea]

@billmetangmo Done on our side (@Joyce-Tchamdjou )

[billmetangmo]

One output key expected, got dict_keys(['result', 'source_documents'])