0.0.3 - 2024-05-28
- Compute and discount Badput from first step after start or restart.
- Compute and discount Badput due to anomalous step times (Pathways only).
- Badput recording APIs
- Some Badput computation APIs (TPU initialization , training preparation, data loading, program startup)
- Goodput monitoring API to asynchronously query and upload Goodput to Tensorboard.
- Bug Fixes
- Fix Goodput calculation with disruptions
- Fix some Cloud Logging latency and batching issues.
0.0.2 - 2024-02-29
- Bug Fixes
- Fixes a typing mismatch in total step time calculation.
- Code and documentation cleanup
0.0.1 - 2024-02-26
- Initial release of ML Goodput Measurement PyPi package
- Feature: Contains the Goodput module which allows logging and retrieval of training job's overall productive Goodput