🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4149

• 📄 Preview Python docs built from this PR

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

⏳ No Failures, 9 Pending

As of commit 2fb716c with merge base bca91b0 ():
💚 Looks good so far! There are no failures yet. 💚

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

@claude review

Claude finished @jerryzh168's task in 2m 1s —— View job

Review of PR #4149: Delete deprecated Int4CPULayout and related code

• Gather context and read the diff
• Review deleted Int4CPULayout implementation
• Check all import/reference removals are complete
• Verify no remaining references to deleted code
• Post review feedback

Overall: Clean deprecation removal. All references to Int4CPULayout, Int4CPUAQTTensorImpl, and the associated dispatch functions have been properly removed across 8 files with no remaining dangling references in the codebase.

Changes Summary

Potential Issue: Leftover config.version = 1 assignments

After this PR, both test/prototype/test_parq.py:346 and torchao/prototype/parq/quant/config_torchao.py:128 still set config.version = 1 when check_cpu_version(device) is true:

if check_cpu_version(device):
    config.version = 1  # <-- still here

However, _int4_weight_only_quantize_tensor at torchao/quantization/quant_api.py:755 has:

assert config.version == 2

This means on CPU with PyTorch >= 2.6, these code paths will hit an assertion error at runtime. This may be intentional (the Int4CPULayout was the only version 1 implementation for int4, so version 1 is now dead), but if so, the config.version = 1 assignments and the check_cpu_version branches should also be removed rather than left as dead code that will fail at runtime.

If this is expected to be cleaned up in a subsequent PR in this ghstack (e.g., one of #4150–#4153), this is fine. Otherwise, consider removing these leftover version 1 assignments in this PR.

Everything else looks good

• No remaining references to Int4CPULayout, Int4CPUAQTTensorImpl, or _linear_fp_act_uint4_weight_cpu anywhere in the codebase
• The raw_output check in affine_quantized_tensor.py:288 correctly narrows to just (TensorCoreTiledLayout, PlainLayout)
• The LAYOUT_TO_* dict cleanups in quant_api.py are correct
• The dispatch deregistration in affine_quantized_tensor_ops.py is properly paired with the deletion of the implementation

@claude review again

Claude finished @jerryzh168's task in 1m 19s —— View job

Re-review of PR #4149: Delete deprecated Int4CPULayout and related code

• Gather context and read the diff
• Review all deleted and modified files
• Check for remaining references to deleted code
• Verify correctness of remaining code paths
• Post review feedback

Overall: This PR is clean and ready to merge. All 8 files are correctly updated, and there are zero remaining references to Int4CPULayout, Int4CPUAQTTensorImpl, int4_cpu_layout, or _linear_fp_act_uint4_weight_cpu anywhere in the codebase.

Changes Summary

Previous issue resolved

The previous review flagged leftover config.version = 1 assignments in test_parq.py and config_torchao.py. This has been addressed — both the check_cpu_version imports and the config.version = 1 branches have been fully removed in the current revision.

CI Status

8 of 14 checks passed (Ruff, Build Docs, L4 tests, aarch64 regression tests, PR label check). 6 checks are still in progress or queued (H100 tests, regression tests, MPS tests). No failures.

No issues found. LGTM.

View job