pytorch / pytorch/executorch

Investigate failure when running metal delegated whisper sampler (argmax)

Open
#16,984 0 comments 1 reaction 2 assignees View on GitHub

@larryliu0820 is already working on this.

Since Jan 28, 2026.

module: metal triaged
Dominant language
Python
Stars
5k
Forks
1.2k
Avg merge
2d 10h
Merged PRs (30d)
581

Description

🐛 Describe the bug

Error reported in https://github.com/huggingface/optimum-executorch/pull/208:

E 00:00:13.059103 executorch:et_metal.mm:246] ETMetalShaderLibrary: Failed to compile shader library: program_source:3813:29: error: assigning to 'bfloat' from incompatible type 'float'
tmp_acc_2 = tmp0;
^~~~

E 00:00:13.059124 executorch:et_metal.mm:263] ETMetalShaderLibrary: Library not compiled
E 00:00:13.059126 executorch:et_metal.mm:301] ETMetalShaderLibrary::getKernelFunction: Failed to get pipeline state for 'generated_kernel'
E 00:00:13.059127 executorch:shim_mps.mm:105] aoti_torch_mps_get_kernel_function: Failed to get kernel function 'generated_kernel'
E 00:00:13.059129 executorch:shim_mps.mm:517] aoti_torch_mps_run_command_block: null function handle

This PR disabled the sampler export code path to let runner fallback to C++ CPU sampler.

Versions

main

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.