DMA delivers a fused low-bit MXFP attention kernel with diagonal tiling that achieves significant speedup on B200 GPUs with negligible generation quality loss.
Sageattention: Accurate 8-bit atten- tion for plug-and-play inference acceleration
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Diagonal-Tiled Mixed-Precision Attention for Efficient Low-Bit MXFP Inference
DMA delivers a fused low-bit MXFP attention kernel with diagonal tiling that achieves significant speedup on B200 GPUs with negligible generation quality loss.