Conversation
Add resumable checkpoints to BenchmarkRunRecorder
Resume benchmark runs from checkpoints by default
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Interrupted benchmarks previously lost completed inference and restarted from scratch. Gasbench 0.10.0 makes checkpointing and resume the default for image, video, and audio benchmarks, using the existing run ID and BenchmarkRunRecorder.
This release includes #161 and #162: append-only durable batches, frozen model/evaluator/sample-plan validation, restoration through the existing scoring and parquet paths, and immediate failure on incompatible or corrupt checkpoints. It also includes the audio sample recovery fix. The package version is bumped from 0.9.10 to 0.10.0 on dev.
Validation: the checkpoint/resume changes passed their unit and CI checks. A live staging NVIDIA A10 acceptance test using bmcore's sandbox adapter terminated the first sandbox after 32 committed predictions. Its replacement restored those 32 and completed the remaining 64; all 96 predictions and probabilities exactly matched an uninterrupted control, with no repeated completed inference and an unchanged original checkpoint batch. The GPU test exercised commit 5521899; the only subsequent change in this release is the version bump.
This PR merges dev into main. Modal sandbox replacement is implemented separately in BitMind-AI/bmcore#408; coordinator recovery and finalization/accounting idempotency remain follow-up work.