biondizzle 77cc28cc92 fix: cutlass.Float32 not cutlass.float32_t in fused epilogue ...

Step 1 SiLU validation: PASS
- cute.exp works on register tensors
- SiLU (x / (1+exp(-x))) in registers matches PyTorch reference
- Relative error: 0.034%, Max abs error: 0.0625 (BF16 precision limit)

2026-05-20 03:12:23 +00:00

..

blockscaled_gemm

refactor: copy CuTeDSL kernel into repo with local imports

2026-05-16 02:57:54 +00:00

moe

fix: cutlass.Float32 not cutlass.float32_t in fused epilogue

2026-05-20 03:12:23 +00:00

__init__.py

refactor: copy CuTeDSL kernel into repo with local imports

2026-05-16 02:57:54 +00:00