fix v7x failure on fractional eval batch test#4572
Conversation
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
08c5710 to
22a8236
Compare
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
| always() && | ||
| needs.analyze_code_changes.outputs.run_notebooks == 'true' && | ||
| needs.build_and_upload_maxtext_package.result == 'success' | ||
| if: false # Temporarily disabled — only tpu7x tests are enabled |
There was a problem hiding this comment.
can you link a buganizer here?
There was a problem hiding this comment.
all my changes in ci_pipeline.yml will be reverted. They are updated for running tpu7x-test.
| tpu7x-tests: | ||
| name: TPU7X tests | ||
| needs: [gate_test_run] | ||
| # REVERT_TPU7X: previously gated to: github.ref == 'refs/heads/main' && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') |
There was a problem hiding this comment.
not a huge fan of commented out code - can you link to buganizer instead?
There was a problem hiding this comment.
all my changes in ci_pipeline.yml will be reverted. They are updated for running tpu7x-test in ci test
22a8236 to
22118f7
Compare
4 participants
Description
Currently v7x test on fractional batch size for eval fails due to the contradictions among
max_target_length=2048andnum_devices=8and splash block size being 512. To fix this, we only need to increase seq length to 4096.Tests
See tpu7x run here: https://github.com/AI-Hypercomputer/maxtext/actions/runs/30033315981/job/89295608280
Checklist
Before submitting this PR, please make sure (put X in square brackets):
gemini-reviewlabel.