biondizzle a468f72a0e CUDA graph: Pre-allocate L1 GEMM output buffers in MoE and SharedExpert ...

Pass out= parameter to run_fused_swiglu_grouped_gemm to avoid per-step
torch.zeros() allocation during CUDA graph capture.

2026-06-03 23:17:43 +00:00

..

__init__.py

Restructure: cutedsl/ -> dsv4/ with proper layering

2026-05-21 17:30:44 +00:00

grouped_linear.py

Fix grouped_linear GEMM output buffer shape and extraction

2026-06-03 22:26:40 +00:00

linear.py

CUDA graph: Fix _assemble_scales_single_group swizzle size

2026-06-03 17:02:34 +00:00

mhc.py

CUDA graph: Fix per-step allocations in decode loop

2026-06-03 16:38:35 +00:00

moe.py

CUDA graph: Pre-allocate L1 GEMM output buffers in MoE and SharedExpert

2026-06-03 23:17:43 +00:00

router.py

CRITICAL FIX: runtime activation global scale to prevent E4M3 overflow

2026-06-01 14:21:16 +00:00

shared_expert.py

CUDA graph: Pre-allocate L1 GEMM output buffers in MoE and SharedExpert

2026-06-03 23:17:43 +00:00