casper-hansen / AutoAWQ

AutoAWQ implements the AWQ algorithm for 4-bit quantization with a 2x speedup during inference. Documentation:
https://casper-hansen.github.io/AutoAWQ/
MIT License
1.69k stars 204 forks source link

Version on PyPi doesn't support Python 3.12 #521

Open horsten opened 3 months ago

horsten commented 3 months ago

So I had to install both this and AutoAWQ_kernels from git. That worked, so there seems to be no underlying reason for this.

pythonLoader commented 2 months ago

I just want to ping that it is still happening. Need to install from source in Python 3.12. @casper-hansen any solution soon?

wasertech commented 2 months ago

I was able to build from source on py312 w/ torch+cu124==2.4.0 using my kernels and AWQ branches. Try to build the kernels from source using: https://github.com/casper-hansen/AutoAWQ_kernels/pull/29 Then you can build AWQ from source using: #573