pytorch / ao

PyTorch native quantization and sparsity for training and inference
BSD 3-Clause "New" or "Revised" License
1.6k stars 179 forks source link

pip install torchao cannot get latest versions (only 0.1 and 2 other version in the same level) #1300

Open moreAImore opened 1 week ago

moreAImore commented 1 week ago

Similar to https://github.com/pytorch/ao/issues/1106 I had py 3.12.7 cuda 12.5 toolkit .5

I had to obtain the WHEEL from another person and install it with the wheel. (Application: ComfyUI)

nitinmukesh commented 1 week ago

Same error

(C:\aitools\cv_venv) C:\aitools>pip install torchao==0.5.0 Looking in indexes: https://pypi.org/simple/, https://pypi.ngc.nvidia.com ERROR: Could not find a version that satisfies the requirement torchao==0.5.0 (from versions: 0.0.1, 0.0.3, 0.1) ERROR: No matching distribution found for torchao==0.5.0

nitinmukesh commented 1 week ago

@moreAImore Please could you share the wheel

moreAImore commented 1 week ago

@moreAImore Please could you share the wheel

Yes. torchao-0.6.1+git-cp312-cp312-win_amd64.zip

nitinmukesh commented 1 week ago

Thank you @moreAImore

moreAImore commented 1 week ago

Thank you @moreAImore

Where did you end up using it?

gau-nernst commented 1 week ago

There are pre-built Windows wheels for nightly build. This should work on windows I think (haven't tested)

pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cpu

Otherwise you can also install from source

USE_CPP=0 pip install git+https://github.com/pytorch/ao

USE_CPP=0 means without CUDA extensions -> less chance for error. If you feel adventurous, you can leave that out to build the CUDA extensions.

nitinmukesh commented 6 days ago

@moreAImore

I actually build from source. I needed it for CogVideo.

torchao 0.7.0+gitd4ca98f6 C:\ai\CogVideo\ao

Wurzeldieb commented 5 days ago

@moreAImore

I actually build from source. I needed it for CogVideo.

torchao 0.7.0+gitd4ca98f6 C:\ai\CogVideo\ao

what pytorch version are you using? 2.5.1+cu124 is giving me an error