Please Fix FP8 Scaled vesions, size in KB.

#7
by krigeta - opened

Hey Team, please fix the scaled versions as the safetensors files that are uploaded are in KB.

Tencent org

Hi, @krigeta , this is expected, as the "scale" safetensor file only contains the necessary scale factor for each parameter for quantization rather than the fp8 model weights.
The fp8 model weights is saved in a separate file.

Thanks for the clarification.

krigeta changed discussion status to closed

Sign up or log in to comment