[Cute,Fwd,Sm100] Fix FP8 precision loss at tile boundaries#2578
Open
Robin930 wants to merge 1 commit into
Open
[Cute,Fwd,Sm100] Fix FP8 precision loss at tile boundaries#2578Robin930 wants to merge 1 commit into
Robin930 wants to merge 1 commit into
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #2577
Problem
FP8 forward pass uses
rescale_threshold=4.0, which combined withmax_offset=8allows intermediate P values up to 2^12 = 4096, exceeding FP8 e4m3 max (448). This causes severe precision loss (cosine similarity drops to ~0.86) at KV tile boundaries (multiples of 128).Fix
Set
rescale_threshold=0.0for FP8 (always rescale O-accumulator). The BF16 path retainsrescale_threshold=8.0as before.Test
Paged decode GQA with kv_seqlen=1~1023. Before/after:
