Settle MATH SFT/CISPO through sidecar-owned mlx-rl - #89
Closed
JoshuaPurtell wants to merge 1 commit into
Closed
JoshuaPurtell wants to merge 1 commit into
JoshuaPurtell wants to merge 1 commit into
Conversation
Admit training from projected optimization_algorithms, persist the SFT dataset digest, retry GPU-blocked mlx-rl event polls, and read CISPO checkpoint identity from sidecar item payloads. Co-authored-by: Cursor <cursoragent@cursor.com>
Contributor
Author
|
Superseded by the current internal Workshop release line. The projected optimizer algorithm admission, SFT dataset validation, and bounded sidecar polling are already present there; the remaining mlx-rl item-id checkpoint behavior was ported with a focused regression in merged PR #117. Closing this conflicting public-main PR so it is not mistaken for release authority. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
optimization_algorithms(and the olderalgorithmshandshake).sft.dataset.validated, retry mlx-rl event polls across GPU-blocked HTTP resets, and settle CISPO checkpoint identity from sidecar itemid.This is the MATH-only settlement from the acceptance-run work, ported onto current
main. It is not the Craftax acceptance branch.Test plan
optimizers::sidecar_training::tests::training_ready_reads_projected_optimization_algorithmsoptimizers::kernel::algorithms::cispo::tests::sidecar_checkpoint_item_id_settlesMade with Cursor