Skip to content

Encapsulate lifecycle management and unify metrics logging in MaxTextTrainingEngine - #4738

Open
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_959402235
Open

Encapsulate lifecycle management and unify metrics logging in MaxTextTrainingEngine#4738
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_959402235

Conversation

@copybara-service

Copy link
Copy Markdown
Contributor

Encapsulate lifecycle management and unify metrics logging in MaxTextTrainingEngine

  • Add internal goodput event recording (TPU_INIT and STEP) inside MaxTextTrainingEngine.
  • Support custom input generation via _gen_model_input_fn and filter auxiliary non-metric tensors in fwd_bwd.
  • Encapsulate XLA compiler HLO dump uploading, profiler lifecycle, step timing, and TFLOPS recording internally within MaxTextTrainingEngine.
  • Unify setup and runtime step metrics logging under metrics_module.MetricsLogger, removing dependency on legacy MetricLogger.

…TrainingEngine

- Add internal goodput event recording (TPU_INIT and STEP) inside MaxTextTrainingEngine.
- Support custom input generation via _gen_model_input_fn and filter auxiliary non-metric tensors in fwd_bwd.
- Encapsulate XLA compiler HLO dump uploading, profiler lifecycle, step timing, and TFLOPS recording internally within MaxTextTrainingEngine.
- Unify setup and runtime step metrics logging under metrics_module.MetricsLogger, removing dependency on legacy MetricLogger.

PiperOrigin-RevId: 959402235
@codecov

codecov Bot commented Aug 5, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant