mlcommons / mlcommons/algorithmic-efficiency

Split out train and update_batch_norm in Librispeech workloads

Open
#163 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
425
Forks
78
PR merge metrics
No merged PRs in 30d

Description

Currently update_batch_norm just runs the librispeech workloads in train mode, which also runs dropout in train mode. The purpose of having separate mode and update_batch_norm kwargs to model_fn() was so that submitters could separate which they want to update, if desired. We can update Conformer.__call__ and Deepspeech.__call__ to take both train and update_batch_norm and pass them to dropout/BN respectively.

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 with algorithmic_efficiency/workloads/librispeech_conformer/librispeech_jax/models.py at Conformer.call and librispeech_deepspeech/librispeech_jax/models.py at Deepspeech.call. Review how model_fn passes mode and update_batch_norm, then verify that train controls dropout while update_batch_norm controls batch normalization independently. Done means both Librispeech workloads support the separated behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.