biondizzle e678afcde0 Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong ...

Key fixes:
- PipelineUmmaAsync consumer group: 32*4=128 threads (not 4 warps)
- TMEM offsets computed from find_tmem_tensor_col_offset (not hardcoded)
- P fragment from p_tmem_s.outer + make_fragment_A (matching fmha.py)
- V SMEM aliasing via recast_ptr

Status:
- Stage A: cosine 0.999999 ✅
- Stage B: runs without crash, identity softmax cosine -0.02 ❌
- Diagnostics: TMEM layout inspection, bisection results

2026-05-20 20:26:25 +00:00

..

kernel

Fused SwiGLU epilogue with granularity-8 weight interleave

2026-05-20 04:13:52 +00:00

kernels

Fix warmup compilation + add sparse topk metadata kernels

2026-05-20 06:43:43 +00:00

__init__.py

refactor: copy CuTeDSL kernel into repo with local imports

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

blackwell_attention.py

Add blackwell_attention module and comprehensive test

2026-05-19 15:30:29 +00:00

bridge.py

stuff

2026-05-20 07:15:01 +00:00

csa_attention.py

Fix device reference in full_attention_reference

2026-05-19 08:01:31 +00:00

custom_ops.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

fp8_bf16.py

feat: GPU-native SWA + sparse decode attention kernels (CuTeDSL)

2026-05-20 05:46:15 +00:00

inverse_rope.py

Replace DeepGEMM fp8_einsum with CuTeDSL NVFP4 for wo_a (o_proj)

2026-05-19 02:36:30 +00:00

moe_pipeline.py

Custom CUDA kernel for de-interleave plus NVFP4 quantize

2026-05-20 04:39:47 +00:00

native_sparse_decode.py

feat: GPU-native SWA + sparse decode attention kernels (CuTeDSL)

2026-05-20 05:46:15 +00:00

native_swa_decode.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

nvfp4_linear.py

Fix warmup compilation + add sparse topk metadata kernels

2026-05-20 06:43:43 +00:00

runner.py

Fix warmup compilation + add sparse topk metadata kernels

2026-05-20 06:43:43 +00:00

shared_expert_pipeline.py

Fix garbled shared_expert_pipeline.py: imports/class were merged

2026-05-19 07:18:10 +00:00

sparse_topk_metadata.py

Fix warmup compilation + add sparse topk metadata kernels

2026-05-20 06:43:43 +00:00

wo_a_grouped_linear.py

Fix wo_a: scatter each group's data at correct offset in padded buffer

2026-05-19 02:45:57 +00:00