Torch is Lit

Torch is Lit

Torch is Lit 🔥❤️‍🔥 is a powerful Retrieval Augmented Generation (RAG) application that leverages Large Language Models (LLMs) to provide an intelligent and comprehensive solution for developers and researchers working with PyTorch and coding tasks.
Date: April-2024
LlamaIndexVectaraTogether.AIStreamlitPython

Torch is Lit 🔥❤️‍🔥

Torch is Lit 🔥❤️‍🔥 is a powerful Retrieval Augmented Generation (RAG) application that leverages Large Language Models (LLMs) to provide an intelligent and comprehensive solution for developers and researchers working with PyTorch and coding tasks.

Features

  1. Lit Torch

    • A RAG application built with Vectara and LlamaIndex
    • Utilizes PyTorch documentation and examples as the corpus data
    • Allows users to ask natural language queries related to PyTorch
    • Retrieves relevant information from the corpus and generates responses using the LLM

  2. Lit Code

    • A coding-tuned LLM hosted on Together.AI
    • Assists with programming tasks through natural language queries
    • Provides intelligent code suggestions, generation, and debugging support

Installation

Prerequisites - Python and Poetry

  1. Clone the repository:

    git clone https://github.com/rony0000013/torch-is-lit.git

  2. Install the dependencies:

    poetry install

  3. Add stonks worker deployed link in .streamlit/secrets.toml

    VECTARA_CUSTOMER_ID=<YOUR_VECTARA_CUSTOMER_ID>
VECTARA_CORPUS_ID=<YOUR_VECTARA_CORPUS_ID>
VECTARA_API_KEY=<YOUR_VECTARA_API_KEY>
LLAMA_PARSE_API_KEY=<YOUR_LLAMA_PARSE_API_KEY>
TOGETHER_API_KEY=<YOUR_TOGETHER_API_KEY>

  4. Run the app

    poetry run streamlit run ui.py

  5. To Add data to courpus add files in data directory and run

    poetry run python file_index.py

Usage

  1. Lit Torch

    • Run the Lit Torch application and ask queries regarding pytorch docs
    • Ask natural language queries related to PyTorch, and the application will retrieve relevant information from the corpus and generate responses using the LLM.

  2. Lit Code

    • Access the Lit Code interface in the streamlit app
    • Provide natural language queries related to programming tasks
    • Lit Code will generate intelligent code suggestions, assist with code generation, and help with debugging.

License

This project is licensed under the Apache 2 License.

Acknowledgments

  • Vectara for providing the RAG toolkit
  • LlamaIndex for the powerful index and query capabilities
  • Together.AI for hosting the coding-tuned LLM