Rollout update - #5
Open
grassesi wants to merge 14 commits into
Open
Conversation
Storing the steps and converting them via chunk_idx/batch_idx lets a partial rollout be indexed without callers tracking offsets. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The predictors now take the global forecast step and let ModelOutput resolve it, so they stay correct when the output covers only part of the rollout. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The context manager had been absorbed into the comment above it, so intermediate rollout steps still built a graph. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sizing the zero arrays from the batch and stream config instead of indexing preds[0] lets a step that holds no predictions for a stream still be written. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The caller knows which steps its data covers, so the writer no longer has to reconstruct them from the target count and the offset. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Taking the steps from ModelOutput lets a partial rollout be written, at the cost of the first chunk emitting empty entries for its leading offset steps. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Collecting the model call and the write in _process_validation_chunks gives the rollout a single place to be split into chunks. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Each chunk continues from the previous one and is written as it completes; the predictions themselves are still retained for the loss computation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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.
Description
Chunked rollout during validation,, make ModelOutput more self-describing
Issue Number
Is this PR a draft? Mark it as draft.
Checklist before asking for review
./scripts/actions.sh lint./scripts/actions.sh unit-test./scripts/actions.sh integration-testlaunch-slurm.py --time 60