Question About Model Quantization and Pruning Approach

#1
by RO-KO - opened

Hi there,

I came across your impressive project and had a quick question out of curiosity. I noticed that the original model is in bfloat16 (bf16), but the quantized version appears to be in float16 (fp16). I was wondering if pruning was part of your quantization process, or if there's another reason for this change in precision format.

Would you mind sharing a bit more detail about your quantization methodology or overall workflow? I'd really appreciate any insights you could offer.

Thanks in advance, and great work again!

Sign up or log in to comment