Repository navigation

CUDA: Optimize accumulation in mmq for NVFP4 type - #29857

Merged

ORippler merged 3 commits into

ggml-org:masterfrom

kmorennv:optimize_mmq_nvfp4

Oct 5, 2026

Commits on Oct 2, 2026

ggml_cuda: optimize accumulation in mmq_vec_dot_fp4_fp4_mma for better performance
kmorennv
committed
5760e10

Commits on Oct 5, 2026

remove whitespace
kmorennv
committed
819a77e
fix: correct indentation in mma_block_scaled_fp4 loop
kmorennv
committed
d4a835e