COMMITS
May 30, 2026
K
[ci feat] Support /bot run tests/<dir> to scope CI test runs (#3422)
Ka-Hyun Nam committed
May 29, 2026
B
fix: replace deprecated APIs: cute.make_fragment and cute.core.ThrMma (#3430)
brandonsun committed
T
Add BGMV MoE CUDA kernels for multi-LoRA (#3249)
Taeho Kim committed
E
feat(trace): add check callbacks to trace templates (#3330)
eigen committed
I
hotfix: cross-warp race in checkpointing SSU kernel (mamba) (#3439)
Igor Shovkun committed
O
[Feat] Add num_heads < 128 support for mla decode kernel (#3309)
Observer007 committed
J
Add mHC post mapping and pre big-fuse kernels (#3285)
jmydurant committed
May 28, 2026
D
feat(kda): add recurrent KDA decode kernel with per-K gating (#2572)
Duncan Moss committed
B
bench: Unify PDL behavior, add missing norm routines, and misc improvements (#3435)
Brian K. Ryu committed
A
feat(autotuner): enable per-op autotune bypass for faster framework warmup (#3396)
Albert Cheng committed
L
test: align test_fmha_v2_prefill SM gating with is_sm12x_supported (#3182)
leonardHONG committed
May 27, 2026
B
Enable smaller tile N for SM100 Cute-DSL NVFP4 GEMM (#3403)
Brayden Zhong committed
Y
Extend autotuner delay kernel length (#3373)
yanqinz2 committed
Z
Make cute dsl mxfp8/nvfp4 quantizer bitwise exact (#3387)
Ziang Li committed
May 26, 2026
W
profiler: group perfetto traces by SM, one row per block (#3038)
Wenxuan Tan committed
V
MLA Decode Autotuning Across TRTLLM-Gen and CuTe Backends (#3355)
Vincent committed
B
S
feat: MNNVL Allreduce quant fusion and performance optimization (#3385)
Shiyu Li committed
B
Enable compression of GPU device binaries (#2949)
Ben Barsdell committed
May 25, 2026
B
Fix returning reference to temporary in moe gemm (#3332)
Ben Barsdell committed
May 22, 2026
R
A
J
Run high-likelihood OOM culprits separately, record memory usage and test duration for analysis (#2961)
Jonathan Dierksen committed
A
bump version to 0.6.12 (#3388)
Alex Yang committed
L
Replace SM120 W4A16 MoE kernels (#3336)
Luke Alonso committed
S
Add torch.compile-compatible custom op for fp4_quantize (#3081)
Serge Panev committed
May 21, 2026