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 2026-05-22 22:00:56 +00:00

12c166245d Use kt directly as TMA GMEM coordinate

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:59:55 +00:00

8c9d4eb1ef Try Int32(0) + kv_coord += 1 (matching reference pattern)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:58:44 +00:00

7e832aa527 Use kvh.count for GMEM tile coordinate (pipeline-tracked SSA value)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:56:57 +00:00

254f7be884 Fix diag: remove .rank

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:56:24 +00:00

4c1dbfd0f3 Add cute.printf shape diagnostics to example9

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:55:43 +00:00

a476324682 Fix TMA shape diag: use ct tensors for LayoutEnum

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:55:05 +00:00

fd6b1e82d8 TMA shape diag: pure Python, no JIT

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:54:24 +00:00

2f670e33d1 TMA shape diagnostic: exact setup from example9 + shape prints

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:53:25 +00:00

b64227e5b6 Fix group_modes range in TMA shape diag

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:52:57 +00:00

be27720cb2 Add TMA shape diagnostic

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:51:35 +00:00

845ad98b22 Fix TMA indexing: 4-mode tensors, kt at mode 2 (GMEM tile dim)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:50:40 +00:00

61b0501a8b Fix test_fmha_v3_stage_c.py: 8-mode TMA indexing + O rescale (from example9)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:43:12 +00:00

0996ffc1ba Add fmha_v3_stage_c_example10: 8-mode TMA + O rescale + paired-atom epilogue

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:36:25 +00:00

328f9b0080 Test n=384

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:36:02 +00:00

2da0a452d1 Quick test n=128,256

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:35:41 +00:00

0d3caced47 Add: O rescale (correction_rescale) in softmax loop + remove pk from TMA/MMA

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:32:19 +00:00

c47d229e6a Sweep test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:31:21 +00:00

a751b3baf7 Sweep test: n=128,256,384,512,1024

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:29:05 +00:00

beaf60db5c DOCUMENT: TMA 8-mode indexing — the bug that cost us a full day. README + inline comments.

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-22 21:21:54 +00:00

27116110ab Fix identity diag: same 8D TMA indexing fix