pytorch / pytorch/executorch

ExecuTorch kernels for native_batch_norm ops do not support mixing float16/bfloat16 input and float32 params

Open
#13,586 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

🕵️‍♂️ Detected with FACTO

In torch, native_batch_norm ops support float16/bfloat16 input, even when params are float32. However, ExecuTorch kernels requires all dtypes to be the same. Expand the support of ExecuTorch kernels to match torch's.

cc @larryliu0820

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.

Research direction

Start by locating the ExecuTorch kernels for the native_batch_norm ops and compare their dtype handling with PyTorch's behavior. Use the FACTO-detected mixed float16/bfloat16 input with float32 parameters as the reproduction case; done means the kernels support the mixed dtypes rather than requiring every value to share one dtype.

Written by the indexing model from the issue text.

Assessment

Tech stack
pytorch
Domain
embedded-iot, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.