llama.cpp
CUDA: fix FA FP16 accumulator overflow for Granite
#18614

Merged

CUDA: fix FA FP16 accumulator overflow for Granite #18614

JohannesGaessler merged 1 commit into ggml-org:master from JohannesGaessler:cuda-fa-fix-overflow

CUDA: fix FA FP16 accumulator overflow for Granite

1d875330

ggerganov approved these changes on 2026-01-05

github-actions added Nvidia GPU

github-actions added ggml

JohannesGaessler merged df17a4c9 into master 51 days ago

Reviewers

ggerganov

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone