biondizzle

0 Followers · 0 Following

• Joined on 2025-12-10

Repositories

25

Projects Packages Public Activity Starred Repositories

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 16:56:10 +00:00

212fc85627 P6: One-way TMEM→regs→SMEM→TMA store epilogue

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:54:24 +00:00

05b5bf9db1 docs: mark P5 as done in NEXT_PRIORITIES.md

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:49:14 +00:00

95e0c8c464 P5: fix multi-tile test — use same Q data for kernel and reference

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:47:01 +00:00

e701a1411c P5: use multi-tile kernel for N>128 in integration test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:46:16 +00:00

5932e928a8 cleanup: remove debug test files (P4, P5)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:45:03 +00:00

8fef46ce73 P5: add reference comparison to Python multi-tile test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:43:30 +00:00

897a70a491 P5: minimal Python multi-tile test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:40:02 +00:00

a2627359fb P5: fix TMA desc creation — write to HOST then cudaMemcpy to device

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:38:35 +00:00

f370bfb1f1 P5: re-enable multi-tile Python tests, fix CAPI to use create_tma_desc_2d_bf16

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:35:36 +00:00

da54f6439f P5: fix TMA multitile test (include cuda.h first, proper SMEM calc)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:34:23 +00:00

34320653e9 P5: standalone TMA multi-tile test with 128B-aligned memory

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:32:46 +00:00

a1d05b3055 P5: disable multi-tile Python tests (TMA descriptor alignment issue)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:30:42 +00:00

97531a68e6 fix: remove n_kv_tiles from capi too

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:28:39 +00:00

a5b47602b5 fix: remove n_kv_tiles from standalone test (struct doesn't have it anymore)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 10:27:40 +00:00

f032800eaa P5: integrate WORKING multi-tile kernel (fmha_6warp_tma_multirow_multitile) into production

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 09:07:10 +00:00

032cb4c7b2 P5: add single-tile merge comparison to multitile test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 09:04:50 +00:00

d424ccbcc1 fix: const not constexpr for SCALE

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 09:04:07 +00:00

3da31de4c0 P5: fix BF16 host helpers for standalone test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 09:01:53 +00:00

9e6ba25a98 P5: standalone multi-tile CUDA test (2 KV tiles, hd=64)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-30 08:59:53 +00:00

b61df2657b P5: fix reference attention for MQA/GQA (kv_idx = h // q_per_kv)