Skip to content

Restore metadata-aware KV checkpoints through PMT - #6679

Open
q10 wants to merge 1 commit into
pytorch:mainfrom
q10:export-D127250013
Open

q10 wants to merge 1 commit into
pytorch:mainfrom
q10:export-D127250013

Conversation

@q10

@q10 q10 commented Oct 30, 2026

Copy link
Copy Markdown
Contributor

Summary: Mark whole-row checkpoint tensors as metadata-prefixed and route restores through the public PartiallyMaterializedTensor.set_weights_and_ids API. Preserve the metadata flag across serialization and select the explicit prefixed-row write path. Materialize checkpoint-view rows in the integration test copy helpers so embedded IDs match the checkpoint-view IDs supplied to restore.

Reviewed By: EddyLXJ

Differential Revision: D127250013

Summary: Mark whole-row checkpoint tensors as metadata-prefixed and route restores through the public `PartiallyMaterializedTensor.set_weights_and_ids` API. Preserve the metadata flag across serialization and select the explicit prefixed-row write path. Materialize checkpoint-view rows in the integration test copy helpers so embedded IDs match the checkpoint-view IDs supplied to restore.

Reviewed By: EddyLXJ

Differential Revision: D127250013
@meta-cla meta-cla Bot added the cla signed label Oct 30, 2026
@meta-codesync

meta-codesync Bot commented Oct 30, 2026

Copy link
Copy Markdown
Contributor

@q10 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D127250013.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant