Repository navigation

[FEAT][kernels] Fused linear log-prob without materializing logits — SM90 TMA + tensor-core forward, chunked backward - #122

Merged

Flink-ddd merged 9 commits into

RL-Align:mainfrom

KJLdefeated:feat/fused-linear-logp

Jun 22, 2026

Merged

[FEAT][kernels] Fused linear log-prob without materializing logits — SM90 TMA + tensor-core forward, chunked backward#122
Flink-ddd merged 9 commits into
RL-Align:mainfrom
KJLdefeated:feat/fused-linear-logp

Commits on Jun 14, 2026
- [FEAT][kernels]: add fused linear log-prob operator (Triton + chunked backward + SM90 TMA/WGMMA)
  
  KJLdefeated
  and
  claude
  committed
  2220177
Commits on Jun 15, 2026
- [FEAT][kernels] SM90 fused linear log-prob: TMA + tensor-core forward, split-V
  KJLdefeated
  committed
  480388e
- Fix the docs
  KJLdefeated
  committed
  6c366c6
- style: black (E305 blank lines)
  KJLdefeated
  committed
  3138dbb
- style: trim trailing whitespace
  KJLdefeated
  committed
  01c53b2
Commits on Jun 16, 2026
- (bug) add target/bias validation
  KJLdefeated
  committed
  bc93e36
Commits on Jun 17, 2026
- check lm_head_weight and hidden device
  KJLdefeated
  committed
  dc3cd73
Commits on Jun 22, 2026
- fix the requests
  KJLdefeated
  committed
  e7966e4
- Merge branch 'main' into feat/fused-linear-logp
  KJLdefeated
  authored
  649132f