Guanchu Wang

icon indicating a website link https://guanchuwang.github.io/home/

icon company Rice University

Results 12 issues of Guanchu Wang

When runing AWQ for WizardLM/WizardLM-7B-V1.0 model, Line 139 of auto_scale.py: "scales = scales / (scales.max() * scales.min()).sqrt()" will lead scales to be inf.

When runing AWQ for WizardLM/WizardLM-7B-V1.0 model, Line 139 of auto_scale.py: "scales = scales / (scales.max() * scales.min()).sqrt()" will lead scales to be inf. Can we revise it into "scales =...

Could you share the code of round-to-nearest?

The results in the paper are amazing! This work will be a milestone in LLM! Recently, our group concentrated on reproducing the experiment results in this work for follow-up research....