Chatllama

A web UI for my fine tuned LLAMA2 and Stable Diffusion model with RAG and voice prompt capabilities.

Update 2

Added the Stable Diffusion XL model fine-tuned using Dreambooth. Type <GEN> before prompt to generate image!

Update!

ChatLLAMA now supports Retrieval Augmented Generation (RAG) via Langchain. Click on Upload file to use a document.

Overview

This repository contains the fine tuned version of llama-2-7b-chat model trained on my dataset with a web interface. Converts voice prompts to text via OpenAI's Whisper model. Model trained using QLoRA and deployed using HuggingFace ecosystem. Frontend in react and served using Flask.

Installation

1. Run
   

git clone https://github.com/mohitpg/chatllama.git

2. Run
   

 python -m venv <path>

3. Activate
   Bash source <path>/bin/activate
   Windows <path>\Scripts\activate.bat
   
   
4. Run
   

pip install -r requirements.txt

5. Go to chatllama/front and run
   

npm run build

6. Login with huggingface-cli
7. Finally run
   

python server.py

This will run the server at localhost port:5000

References

https://arxiv.org/abs/2305.14314
https://arxiv.org/abs/2402.18158
https://arxiv.org/abs/2208.12242