Repository navigation

[CUDA] Optimize HyperConnectionLowRankMix for S<=8 speculative decode and deterministic prefill - #33548

Open

Akshay Sonawane (apsonawane) wants to merge 15 commits into

mainfrom

asonawane/dflash2-s4-hyperconnection

Open

[CUDA] Optimize HyperConnectionLowRankMix for S<=8 speculative decode and deterministic prefill#33548
Akshay Sonawane (apsonawane) wants to merge 15 commits into
mainfrom
asonawane/dflash2-s4-hyperconnection

Commits on Oct 23, 2026
- [CUDA] Tune HyperConnectionLowRankMix 4-row decode, rank-320 specialization, and deterministic prefill post-gate
  
  Tianlei Wu (tianleiwu)
  and
  Copilot
  committed
  1428ea7
Commits on Oct 24, 2026
Commits on Oct 26, 2026