BLASST wins Best Paper at MLSys 2026 for a drop-in training-free dynamic sparse attention mechanism that thresholds online softmax statistics to skip negligible blocks in long-context LLM inference

— It targets self-attention compute and memory bottlenecks during inference.

Original post

SH#1330@SONGHAN_MITOPHMHuizi Mao|@HUIZI_MAO

Glad to be featured by SemiAnalysis. Our work BLASST was also selected as MLSys 2026 Best Paper: https://mlsys.org/virtual/2026/poster/3631

3:35 PM · May 17, 2026 View on X

Reposted by

DZ#1209|@ZDHNARSIL

QUOTE POSTFI#397finbarr|@FINBARRTIMBERS

This is an elegant paper; hope to try it out soon.

1:42 PM · May 18, 2026 · 15.8K Views