biondizzle b3d1aae038 feat: full FP4 activations for mxf4nvf4 - E2M1 packed A side + UE4M3 scales ...

mxf4nvf4 requires BOTH A and B to be FP4 (E2M1 packed).
Changes:
- a_dtype_t: float_e4m3_t → float_e2m1_unpacksmem_t
- UMMA_K: 32 → 64 (FP4 MMA atom)
- L1 epilogue: FP8 quant → E2M1 FP4 quantization with nearest-neighbor
- L1 output SMEM: packed E2M1 (2 per byte), TMA store uint8
- TMA descriptors: adjusted for FP4 packing (K/2 bytes per row)
- SymmBuffer: uint8 activations, shape (M, K//2)
- Staging kernel: BF16 → E2M1 packed + UE4M3 block16 scales

2026-05-11 20:29:08 +00:00

..

apis

feat: full FP4 activations for mxf4nvf4 - E2M1 packed A side + UE4M3 scales

2026-05-11 20:29:08 +00:00

indexing

[Public release 26/04] Introducing Mega MoE, FP4 Indexer and other features/fixes (#304)

2026-04-17 09:45:14 +08:00

jit

fix: gran_k=16 in transform_sf + sm_100a arch for NVFP4 mega_moe

2026-05-11 16:11:11 +00:00

jit_kernels

feat: full FP4 activations for mxf4nvf4 - E2M1 packed A side + UE4M3 scales

2026-05-11 20:29:08 +00:00

utils

[Public release 26/04] Introducing Mega MoE, FP4 Indexer and other features/fixes (#304)

2026-04-17 09:45:14 +08:00

python_api.cpp

feat: NVFP4 mega MoE kernel (scale_vec::4X, UE4M3 block scales)

2026-05-11 05:41:08 +00:00