does it make sense to have int8 for ampere ? could you share the quantization script ?
#5
by
erosdiffusion
- opened
First of all, thanks for the amazing job (as always top quality work).
a small question / request:
as far as I know int8 is good for ampere cards (3080) as it's hardware supported.
If that's true, would it be possible to make an int8 version ?
Or... could you share the quantization script so I can try myself ?
erosdiffusion
changed discussion title from
does it make sense to have int8 for ampere ?
to does it make sense to have int8 for ampere ? could you share the quantization script ?