This is small LLM project using Langchain with RAG and ReAct acting as a Doctor. The RAG use MASHQA dataset as the reference for the model to be better at answering questions.
- Model still sometime doesn't follow the rules set by ReAct prompt
- When using with ReAct, RAG sometimes only return question from the vectordb instead of both question and answer. Which result into the final answer being a question instead of an answer toward the user.
- ReAct doesn't work well with smaller model.
- langchain
- langchain-community
- langchainhub
- langchain-openai
- langchain-huggingface
- bs4
- streamlit
- faiss-cpu
pip install -r requirements.txt
streamlit run app.py