Repository navigation

[Deferred] Fused Triton GDN/KDA decode QAT - #2562

Draft

kaix-nv wants to merge 6 commits into

kaix/linear-attention-qat-examplefrom

kaix/linear-attention-decode-triton

Draft

[Deferred] Fused Triton GDN/KDA decode QAT#2562
kaix-nv wants to merge 6 commits into
kaix/linear-attention-qat-examplefrom
kaix/linear-attention-decode-triton

Commits on Sep 22, 2026
- Add GDN state and WY activation QAT emulation
  kaix-nv
  committed
  07d39f3
Commits on Sep 28, 2026
- Qualify GDN QAT on Hopper and Megatron training
  kaix-nv
  committed
  bfbb13e
Commits on Sep 29, 2026
- Move GDN test compilation into setup fixtures
  kaix-nv
  committed
  95766de
- Clarify dynamic per-tile FP8 state quantization
  kaix-nv
  committed
  3896181
- Add GDN/KDA Torch decode QAT with INT8 recurrent state
  kaix-nv
  committed
  d03c732
- Add fused Triton GDN/KDA decode QAT
  kaix-nv
  committed
  756c330