coding-crashkurse/langchain-production-project

langchain-production-project's Issues

Service3/app.py Edits

Hey pretty cool work.

I reckon format_docs function in service3/app.py might need a bit of change. It just takes in the source of docs and no page content leading to model making hallucinating. So adding doc.page_content should solve it.

Cheers

History for RetrievalQA

Hi, I was trying this setup for RetrievalQA but I'm unable to fetch history for the same. This is how I have updated the service 3. I have utilized deeplake local vector db for the same.
requirements
langchain==0.0.201 openai==0.27.8 unstructured==0.7.5 deeplake==3.6.4 tiktoken==0.4.0 python-dotenv

@app.post("/service3/{conversation_id}")
async def service3(conversation_id: str, conversation: Conversation):
print('Inside service3')
print(conversation)
query = conversation.conversation[-1].content

deeplake_outdir = "./VectorDB/IQA-v2-2docs/"

texts = pdf_loder()
add_data_vector_db(texts, deeplake_outdir)
dl_db = read_vector_db(deeplake_outdir)
retriever = dl_db.as_retriever()
prompt_template = """ Following is a legal statement of work (SOW) for a contract. You are an intelligent AI which analyses contract snippets and answers the questions. If you don't know the answer, just say that you don't know, don't try to make up an answer. 
{context}
Question: {question} 
Answer:
"""
PROMPT = PromptTemplate(template=prompt_template, input_variables=["context", "question"] ) 

chain_type_kwargs = {"prompt": PROMPT}
model = AzureChatOpenAI(
    # model_name = "gpt-4",
    model_name = "gpt-3.5-turbo",
    openai_api_base=os.environ['OPENAI_API_BASE'],
    openai_api_version=os.environ['OPENAI_API_VERSION'],
    deployment_name="",
    openai_api_key=os.environ["OPENAI_API_KEY"],
    openai_api_type = "azure",
)
qa = RetrievalQA.from_chain_type(llm = model, chain_type='stuff', retriever=retriever, chain_type_kwargs=chain_type_kwargs,return_source_documents=True)

response = qa({"query": query}, return_only_outputs=True)
return {"id": conversation_id, "reply": response['result']}`

When I ask 2-3 questions to this and I ask let's say what was my first question the model is returning the complete context and not the history. Also, in my redis I can see the Human and Assitant content getting stored. Any help would be appreciated around working this for RetrievalQA.