nightly build for mac stops on 0422

[guangy10]

For example running pip install torchao==0.11.0.dev20250430 --extra-index-url "https://download.pytorch.org/whl/nightly/cpu" will not work on mac. The latest build available on mac was from 0422.

[guangy10]

cc: @jerryzh168

[atalman]

Something changed on 0419
From https://download.pytorch.org/whl/nightly/torchao/

torchao-0.11.0.dev20250416+cpu-py3-none-any.whl
torchao-0.11.0.dev20250417+cpu-py3-none-any.whl
torchao-0.11.0.dev20250418+cpu-py3-none-any.whl
torchao-0.11.0.dev20250419+cpu-cp39-abi3-manylinux_2_28_x86_64.whl
torchao-0.11.0.dev20250420+cpu-cp39-abi3-manylinux_2_28_x86_64.whl
torchao-0.11.0.dev20250421+cpu-cp39-abi3-manylinux_2_28_x86_64.whl
torchao-0.11.0.dev20250422+cpu-py3-none-any.whl
torchao-0.11.0.dev20250423+cpu-cp39-abi3-manylinux_2_28_x86_64.whl
torchao-0.11.0.dev20250424+cpu-cp39-abi3-manylinux_2_28_x86_64.whl
torchao-0.11.0.dev20250425+cpu-cp39-abi3-manylinux_2_28_x86_64.whl

[atalman]

Debugging this:
This run builds good wheel: https://github.com/pytorch/ao/actions/runs/14530034363
This is broken: https://github.com/pytorch/ao/actions/runs/14530210383?pr=1372

[atalman]

This is the culprit PR that caused this regression on April 19: #1372
It was reverted on April 22 : #2092 hence we where able to generate good binary
Here: https://github.com/pytorch/ao/actions/runs/14583749783/job/40905435628

However next PR that broke is this #2044
Run : https://github.com/pytorch/ao/actions/runs/14584443629/job/40907370225

[guangy10]

@atalman Is this fixed? Running pip install torchao==0.11.0.dev20250506 --extra-index-url "https://download.pytorch.org/whl/nightly/cpu" on mac will still end up with the same issue:

Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/nightly/cpu
ERROR: Could not find a version that satisfies the requirement torchao==0.11.0.dev20250505 (from versions: 0.0.1, 0.0.3, 0.1, 0.5.0, 0.6.1, 0.7.0, 0.8.0, 0.9.0, 0.10.0.dev20250302, 0.10.0.dev20250302+cpu, 0.10.0.dev20250303, 0.10.0.dev20250303+cpu, 0.10.0.dev20250304, 0.10.0.dev20250304+cpu, 0.10.0.dev20250305, 0.10.0.dev20250305+cpu, 0.10.0.dev20250306+cpu, 0.10.0.dev20250307+cpu, 0.10.0.dev20250308+cpu, 0.10.0.dev20250309+cpu, 0.10.0.dev20250310+cpu, 0.10.0.dev20250311+cpu, 0.10.0.dev20250312+cpu, 0.10.0.dev20250313+cpu, 0.10.0.dev20250314+cpu, 0.10.0.dev20250315+cpu, 0.10.0.dev20250316+cpu, 0.10.0.dev20250317+cpu, 0.10.0.dev20250318+cpu, 0.10.0.dev20250319+cpu, 0.10.0.dev20250320+cpu, 0.10.0.dev20250321+cpu, 0.10.0.dev20250322+cpu, 0.10.0.dev20250323+cpu, 0.10.0.dev20250324+cpu, 0.10.0.dev20250325+cpu, 0.10.0.dev20250326+cpu, 0.10.0.dev20250327+cpu, 0.10.0.dev20250328+cpu, 0.10.0.dev20250329+cpu, 0.10.0.dev20250330+cpu, 0.10.0.dev20250331+cpu, 0.10.0.dev20250401+cpu, 0.10.0, 0.10.0+cpu, 0.11.0.dev20250403+cpu, 0.11.0.dev20250404+cpu, 0.11.0.dev20250411+cpu, 0.11.0.dev20250412+cpu, 0.11.0.dev20250413+cpu, 0.11.0.dev20250414+cpu, 0.11.0.dev20250415+cpu, 0.11.0.dev20250416+cpu, 0.11.0.dev20250417+cpu, 0.11.0.dev20250418+cpu, 0.11.0.dev20250422+cpu)
ERROR: No matching distribution found for torchao==0.11.0.dev20250506

[Valentine233]

@atalman @jerryzh168 Could you please describe what building issue #2093 causes? I haven't see any related logs here.

[atalman]

@guangy10 Yes fixed take a look at: https://download.pytorch.org/whl/nightly/torchao/
Since its a nightly build you need to wait for next morning to see the result:

[atalman]

HI @Valentine233 the Original PR caused same regression April 19: #1372

Please refer to this comment:
#2157 (comment)

Issue is that this PR generates torchao-0.12.0.dev20250507+cpu-cp39-abi3-manylinux_2_28_x86_64.whl (notice manylinux) Please see the PR that tries to reland this code: #2183

[guangy10]

@atalman Yeah noticed it. Already moved the pin to latest nightly here: huggingface/optimum-executorch#68. Thanks

[Valentine233]

@atalman I find the xx+cpu-cp39-abi3-manylinux_2_28_x86_64.whl is automatically generated as long as we add cpp files under folder torchao/csrc/. I check cuda and we also use the manylinux wheel: xx+cu118-cp39-abi3-manylinux_2_28_x86_64.manylinux_2_24_x86_64.whl. So how does cuda wheel works on macOS? Can cpu follows the same way as cuda?

Besides, is there any CI for macOS wheel building available to be triggered?

cc @mingfeima @leslie-fang-intel