pytorch / ao

PyTorch native quantization and sparsity for training and inference
BSD 3-Clause "New" or "Revised" License
1.56k stars 168 forks source link

torchao install error from torchchat in Google colab #589

Open sunshinesfbay opened 3 months ago

sunshinesfbay commented 3 months ago

I was following this youtube video's intro to torchchat => https://www.youtube.com/watch?v=bIDQeC0XMQ0 and ran into the same problem the presenter described where torchao fails to build in this colab => https://drive.google.com/file/d/1eut0kyUwN7l5it6iEMpuASb0N33p9Abu/view?usp=sharing

msaroufim commented 3 months ago

So following that example did work well for me

To install ao you'd need to install it in this way

pip install --pre torchao-nightly --index-url https://download.pytorch.org/whl/nightly/cu121 # full options are cpu/cu118/cu121/cu124

I do see a comment on the torchchat side saying they'll fix the installation instructions here https://github.com/pytorch/torchchat/blob/main/install_requirements.sh#L74 cc @Jack-Khuu but this is gated on us providing mac binaries which don't provide as of now. @atalman is this is an easy change we could potentially make?

msaroufim commented 3 months ago

So we now have m1 support for our release binaries, as of today this 0.4 which I pushed manually. Not planning on adding nightly support until we figure out automation

jerryzh168 commented 3 months ago

Also pip install torchao-nightly installs torchao_nightly-2024.5.29-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl for me, can we fix this?