Rerun the conversion

#3
by michaelfeil - opened

FYI we just upgraded our weights. The alignment should be much better now. Thanks for the 2bit quant

Pruna AI org
edited May 6

Thanks for the notification. We will do it asap :)

Pruna AI org

The model is updated ;) Since 2 bits quant is quite aggressive, it might still need some PEFT to recover from it.

sharpenb changed discussion status to closed

Yeah, I think for long context, even int8 might be too agressive

Sign up or log in to comment