Avoid repeated MoE workspace scans in GDN planning - #998
Merged
Merged
Conversation
bradhilton
deployed
to
trainer-rank-gpu-validation
September 26, 2026 21:28 — with 
GitHub Actions
Active
GitHub Actions
Active
bradhilton
added a commit
that referenced
this pull request
Sep 26, 2026
This was referenced Sep 26, 2026
This branch was successfully deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
GDN pending-memory pricing recalculates the same named-slot MoE workspace once per GDN layer, and each calculation walks the model. Compute it once per gradient group, then retain the same maximum over per-layer pending memory. The value is still refreshed for every group and every call; slot metadata, memory floors, and admission checks remain unchanged.
The regression covers 1 and 30 GDN shapes, default and named slots, mixed gradient/reference groups, refreshed prices on later calls, and unsupported/no-gradient paths. All 71 pending-memory tests pass with GPUs hidden; Ruff and diff checks pass. The retained native profile motivated this change, but its GPU utilization and wall-time benefit have not been measured.