This project implements a Retrieval-Augmented Generation (RAG) chatbot designed to answer questions about machine learning papers. It uses a dataset of 100 LLM (Large Language Model) papers to provide informative and contextually relevant responses to user queries about recent advancements in the field of machine learning, particularly focusing on LLMs.
- Retrieval-Augmented Generation (RAG) for accurate responses about ML papers
- Utilizes a dataset of 100 recent LLM papers
- Configurable model parameters and pipeline components
- Efficient PDF processing, document splitting, and embedding
- FAISS vector store for fast similarity search
- Command-line interface for quick queries
- Streamlit web interface for interactive chatting (optional)
This project uses the "100 LLM Papers to Explore" dataset from Kaggle: https://www.kaggle.com/datasets/ruchi798/100-llm-papers-to-explore
The dataset includes 100 PDF papers covering various aspects of Large Language Models, providing a rich knowledge base for the chatbot.
-
Clone the repository:
git clone https://github.com/MahtabRanjbar/ml-papers-rag.git cd ml-papers-rag -
Install the required packages:
pip install -r requirements.txt
-
Configure the project:
- Open
config/config.ymland adjust the settings as needed, especially the path to the downloaded papers
Run a single query:
python src/main.py "What are the recent advancements in LLM fine-tuning techniques?"This project uses the following models:
- LLM: Meta-Llama-3-8B-Instruct
- Embeddings: BAAI/bge-base-en-v1.5
You can change these in the config.yml file.
The chatbot's knowledge base is built from a collection of machine learning papers. You can replace or expand this dataset by adding PDF files to the configured input directory.
All configurable parameters are stored in config/config.yml. This includes:
- Model settings (name, temperature, top_p, etc.)
- Text splitting parameters
- Embedding model selection
- File paths for input and output