Repository navigation
Conversation
shihaobai
force-pushed
the
bsh/glm-ffn-swiglu-quant
branch
from
November 12, 2026 11:12
aaa5cee to
b2d0dba
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fuse GLM dense/shared FFN activation and quantization for BF16 DeepGEMM with unpacked UE8M0 scales. Preserve clipping and intermediate rounding; other formats retain the existing path. The small shared-kernel alpha extension is identical to #1709; this PR adds the separate dense/shared consumer.
Validated on H100 and retested after rebasing onto main db192ec: 211 GPU tests pass, including scale/FP8 parity, fallbacks and CUDA graph replay. With actual GLM checkpoint weights, all 32 FFN comparisons are byte-exact (1–8192 tokens). At 8192 tokens the full dense/shared FFN is 1.25x/1.19x faster; small-batch CUDA graph gains are 1.02–1.11x. These are local FFN timings, not end-to-end TPM.