lmsys

/

vicuna-7b-delta-v1.1

like 202

I understand the difficult situation with LLaMa weights, converting to HF model, and distributing Vicuna as diffs

Would you be open to posting hashes to verify the final Vicuna model? I don't think this risks original values being reverse-engineered. For example:

sha1sum pytorch_model-0000*
167df1a63ff68eba24f19aed3de6f9fe852be460  pytorch_model-00001-of-00002.bin
ff706207fe385b8065a1030d582d9dd193558533  pytorch_model-00002-of-00002.bin