WIP: Enable Parakeet on Metal

[mergennachin]

No description provided.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16499

• 📄 Preview Python docs built from this PR

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

❌ 3 New Failures, 1 Unrelated Failure

As of commit fc45eed with merge base 55fe42b ():

NEW FAILURES - The following jobs have failed:

• pull / test-samsung-models-linux / linux-job (gh)
  test_resnet50_fp16
• Test CUDA Builds / test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command docker exec -t f61dc8ddd63c3c0cb9d86a11ef39849139be0427b4015fdf6e22c2e26e1326c7 /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-small, quantized-int4-weight-only) / linux-job (gh)
  RuntimeError: Command docker exec -t 54f35375075424faba021c29c6535ccdb6357ee74d9d762e7a7152f25746f393 /exec failed with exit code 1

UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:

• pull / android / run-emulator (gh) (#16137)
  Timeout waiting for emulator to boot.

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[mergennachin]

@JacobSzwejbka @manuelcandales @Gasoonjia

Here's the latest update:

1. Export is good.
2. Runtime is failing with https://gist.github.com/mergennachin/a8b5dbcb923a17ac1f84c0dda27bad77
3. aoti_torch_mps_addmm_out implementation was missing too on top of aoti_torch_mps_bmm_out
4. There was a bug with convolution op -- fixed in this PR
5. Fix MPS constant blob offset indexing in AOTI runtime pytorch#171998 seems like a false alarm - closing the PR for now.

[mergennachin]

Here's where the stacktrace during assertion

https://gist.github.com/mergennachin/8a43299f4e74c16da3dd2946275737cf

[manuelcandales]

I have been looking at the inductor generated code, and this is the line from the generated code that causes the error:

AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_mps_addmm_out(buf12, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(pre_encode_out_bias, 2, int_array_10, int_array_12, 0LL)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf11, 2, int_array_13, int_array_14, 0LL)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(pre_encode_out_weight, 2, int_array_15, int_array_16, 0LL)), 1LL, 1LL));

The error is produced by that first call to reinterpret_tensor_wrapper:

reinterpret_tensor_wrapper(pre_encode_out_bias, 2, int_array_10, int_array_12, 0LL))

pre_encode_out_bias contains 1024 elements
int_array_10 = {20, 1024} (these are the sizes)
int_array_12 = {0, 1} (these are the strides)

reinterpret_tensor_wrapper turns into a call to aoti_torch__reinterpret_tensor, which calls executorch::extension::from_blob which calls make_tensor_ptr

So, this turns into a call to make_tensor_ptr with data containing 1024 elements, sizes = {20, 1024} and strides = {0, 1}

[mergennachin]

Closing in lieu of #16562