Repository navigation

[CUDA] Add BFloat16 ArgMax/ArgMin support and extend GatedAdd with projected gate - #33470

Merged

David Fan (jiafatom) merged 21 commits into

mainfrom

asonawane/main-ops-hyperconnection-tensor

Oct 27, 2026

Merged

[CUDA] Add BFloat16 ArgMax/ArgMin support and extend GatedAdd with projected gate#33470
David Fan (jiafatom) merged 21 commits into
mainfrom
asonawane/main-ops-hyperconnection-tensor

Commits on Oct 19, 2026
Commits on Oct 22, 2026
- Add shared-expert preparation and projected sigmoid gates
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  b679f3b
- Merge main into asonawane/main-ops-hyperconnection-tensor
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  dd869d5
- Apply lintrunner formatting to fused tensor operators
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  b442b7c
- Remove trailing whitespace in hyperconnection operator documentation
  
  Akshay Sonawane (apsonawane)
  and
  Copilot
  committed
  dbd8b83
- Extend EmbeddingRMSNorm, RowSelect, and deterministic HyperConnectionLowRankMix post-gate
  
  Tianlei Wu (tianleiwu)
  and
  Copilot
  committed
  07091eb
- Remove HyperConnectionLowRankMix from PR #33470 (covered by PR #33384)
  
  Tianlei Wu (tianleiwu)
  and
  Copilot
  committed
  6c978ad
Commits on Oct 23, 2026
Commits on Oct 24, 2026
- fix(docs): sync SigmoidGateMul and ArgMaxLastAxis schema docs in ContribOperators.md
  
  Tianlei Wu (tianleiwu)
  and
  Copilot
  committed
  6b8e19d
Commits on Oct 25, 2026
Commits on Oct 26, 2026