facebookresearch / facebookresearch/SpinQuant
Question on Activation Quantization
Open
- Dominant language
- Python
- Stars
- 428
- Forks
- 93
- PR merge metrics
- No merged PRs in 30d
Description
Hi, thanks for the great work.
I am curious regarding the specifics of activation quantization.
In particular, I'd be interested in ablating the benefits of grouped per token quantization vs simple per channel quantization accross bitwidths. Could you kindly provide some advice as to how could the ActivationQuantizationWrapper be modified for enabling such an experiment?
Best
Contributor guide
Assessment
This issue has not been assessed yet.