FastChat

Vicuna

pip3 install fschat

# Request LLaMa Weights to meta (https://docs.google.com/forms/d/e/1FAIpQLSfqNECQnMkycAp2jP4Z9TFX0cGR4uf7b_fBxjY_OjhJILlKGA/viewform)

# Transform into Hugging Face format
python src/transformers/models/llama/convert_llama_weights_to_hf.py \
    --input_dir ~/Downloads/llama/weights --model_size 7B --output_dir ~/.llama-13b

# Delta Vicuna Weights
python3 -m fastchat.model.apply_delta \
    --base ~/.llama-13b \
    --target ~/.vicuna-13b \
    --delta lmsys/vicuna-13b-delta-v0

# Serving
python3 -m fastchat.serve.cli --model-name ~/.vicuna-13b

Vicuna: An Open-Source Chatbot Impressing GPT-4 with 90%* ChatGPT Quality

by the Team with members from UC Berkeley, CMU, Stanford, and UC San Diego

https://vicuna.lmsys.org/