Repository navigation
CUDA: fix FP16 overflow in tile FA kernel - #17875
Merged
JohannesGaessler merged 1 commit intoDec 9, 2025
Merged
Conversation
github-actions
Bot
added
Nvidia GPU
ggerganov
approved these changes
Dec 9, 2025
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
my-other-github-account
pushed a commit
to my-other-github-account/llama.cpp
that referenced
this pull request
May 15, 2026
fukuro-kun
pushed a commit
to fukuro-kun/fukuro-llama-cpp-turboquant
that referenced
this pull request
Jul 5, 2026
MrLordCat
referenced
this pull request
in MrLordCat/llama.cpp-rdna-lab
Jul 16, 2026
zommiommy
pushed a commit
to zommiommy/llama.cpp
that referenced
this pull request
Aug 18, 2026
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to #17558 .
I forgot that the tile kernel would also be suffering the same problem with overflows. Unfortunately unlike the vector kernel the tile kernel is compute bound rather than I/O bound so simply switching the arithmetic to FP32 is not a good option. I think the least bad option is to scale down Q and to then scale up KQ afterwards again. The tradeoff is potentially flushing some small values <$2^{-12}$ to zero. Affected GPUs are P100s, V100s, and RDNA1 or older AMD GPUs.