biondizzle e1fcfc4f01 Add CuTeDSL warmup + CUDA sync after JIT compilation ...

CuTeDSL cute.compile corrupts GPU memory. Add warmup forward +
torch.cuda.synchronize() + health check after finalize_weights,
matching the MoE runner pattern.

2026-05-19 01:11:44 +00:00

..

kernels/linear/nvfp4

Add CuTeDSL warmup + CUDA sync after JIT compilation

2026-05-19 01:11:44 +00:00

patches

Register CuTeDSL as proper NvFp4LinearKernel for NVFP4 linear layers

2026-05-19 00:44:44 +00:00

cutedsl_quant_method.py

Fix: add abstract create_weights to CuTeDSLNvfp4LinearMethod

2026-05-18 20:40:48 +00:00

nvfp4_cutedsl.py

Fix torch.compile: use custom autograd Function instead of @torch.compiler.disable

2026-05-18 21:38:28 +00:00