vicuna-7b

The repo contains the converted vicuna-7b model files.

The base model is from decapoda-research/llama-7b-hf and the delta model is from lmsys/vicuna-7b-delta-v0.
The conversion script is

python3 -m fastchat.model.apply_delta \
    --base decapoda-research/llama-7b-hf \
    --target /output/path/to/vicuna-7b \
    --delta lmsys/vicuna-7b-delta-v0

tree -s vicuna-7b

vicuna-7b
├── [ 9976642750]  pytorch_model-00001-of-00002.bin
├── [ 3500323731]  pytorch_model-00002-of-00002.bin
└── [     499723]  tokenizer.model

tree -s vicuna-7b-delta-v0

vicuna-7b-delta-v0/
├── [ 9976642750]  pytorch_model-00001-of-00002.bin
├── [ 3500323731]  pytorch_model-00002-of-00002.bin
├── [      26788]  pytorch_model.bin.index.json
└── [     499723]  tokenizer.model

Although the file size with same prefix is equal, but the content is different. And the md5sum result is

5e63e46f2547e75c209c79aebe21f5ca  vicuna-7b/pytorch_model-00001-of-00002.bin
53a0d7b5e34c432401f1979b454272ee  vicuna-7b/pytorch_model-00002-of-00002.bin

7ded4bec6556250247b1efbb3b7e77e8  vicuna-7b-delta-v0/pytorch_model-00001-of-00002.bin
ee48ac060a0f352ba8fa3dea0385e4c3  vicuna-7b-delta-v0/pytorch_model-00002-of-00002.bin

The converted model has been tested using vicuna cli and got correct response. So enjoy it~

Q: hi how are you
A: I'm just a computer program, so I don't have feelings or emotions like a human. I'm here to respond to your questions and provide information to the best of my ability. Is there anything else you'd like to know?

Q: really cool
A: Thank you! I'm glad you think so. Is there anything in particular you'd like to know or discuss? I'm here to help so feel free to ask me anything.