Repository navigation
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Write GLM sparse MLA Q/V projections directly in token-major layout to avoid the following layout copy. Enable for supported BF16 Hopper shapes in eager Prefill with one TP rank per DP; keep the reference path for short cold inputs and other configurations.
Validation against main dc53088 on H100: 15 GPU tests pass, plus byte-exact checks using checkpoint layers 3/43, two seeds and 1024/1659/4096/8192 rows. At 8K rows, Q takes 0.981 → 0.315 ms and V 0.627 → 0.294 ms. These are projection-plus-layout timings, not system TPM gains.