[CPU] Fix issue: No module named 'fbgemm_gpu.experimental'

[shiyang-weng]

Fix the following issue

  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/transformers/modeling_utils.py", line 50, in <module>
    from torchao.quantization import Int4WeightOnlyConfig
  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/torchao-0.13.0+git38de0e9-py3.10.egg/torchao/__init__.py", line 43, in <module>
    from torchao.quantization import (
  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/torchao-0.13.0+git38de0e9-py3.10.egg/torchao/quantization/__init__.py", line 44, in <module>
    from .quant_api import (
  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/torchao-0.13.0+git38de0e9-py3.10.egg/torchao/quantization/quant_api.py", line 70, in <module>
    from torchao.quantization.quantize_.workflows import (
  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/torchao-0.13.0+git38de0e9-py3.10.egg/torchao/quantization/quantize_/workflows/__init__.py", line 1, in <module>
    from .int4.int4_preshuffled_tensor import (
  File "/opt/conda/envs/pytorch/lib/python3.10/site-packages/torchao-0.13.0+git38de0e9-py3.10.egg/torchao/quantization/quantize_/workflows/int4/int4_preshuffled_tensor.py", line 31, in <module>
    from fbgemm_gpu.experimental.gen_ai.quantize import (
ModuleNotFoundError: No module named 'fbgemm_gpu.experimental'

We want to use torchrec on CPU. On CPU, torchrec dependent on the CPU version of fbgemm_gpu.
On CPU, fbgemm_gpu not include fbgemm_gpu.experimental.

>>> import fbgemm_gpu
>>> fbgemm_gpu.experimental
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'fbgemm_gpu' has no attribute 'experimental'

env: pip install fbgemm-gpu --index-url https://download.pytorch.org/whl/cpu
Reproduce: from torchao.quantization import Int4WeightOnlyConfig

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2591

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 916abda with merge base 12ff479 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[CaoE]

Is there any test for this ?

[shiyang-weng]

Hard to test. If want to test this case, need to install the CPU version of fbgemm on ci envs.

[shiyang-weng]

@pytorchbot label "topic: not user facing"

[shiyang-weng]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here