get_float_qparams 忽略量化粒度？

Open guanchenl opened this issue 7 months ago • 2 comments

get_float_qparams里求出来的tensor和scales形状永远一样，量化粒度：per-tensor/group/channel失效了？

May 25 '25 06:05 guanchenl

Another bug

https://github.com/ModelTC/llmc/blob/main/llmc/compression/quantization/quant.py#L979 起至init函数结束，不该缩进

May 26 '25 01:05 guanchenl

@chengtao-lv

May 26 '25 07:05 gushiqiao