biondizzle

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 10:13:21 +00:00

489c620159 docs: document M_for_layout=128 assumption in _prepack_weight_sf

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 10:12:43 +00:00

b7c7e9fb50 refactor: clean up slot_token handling in cutlass_grouped_nvfp4_gemm

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 10:06:10 +00:00

7a1538d0c8 fix: gather on slot_token presence, add shape asserts L1→L2

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 10:03:38 +00:00

3cc00b12df fix: prepack cache key includes data_ptr, shape, dtype, device, N, K

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 10:00:59 +00:00

3ba41b9322 fix: use slot_token identity check instead of shape heuristic for gather

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:56:48 +00:00

ded80be133 refactor: unify L1/L2 to use 1D slot_expert_ids consistently

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:49:21 +00:00

093babadc6 docs: clarify L1 interleave removal — transpose is still needed

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:42:16 +00:00

c7db2242ee fix: pass slot_expert_ids directly to L2 instead of rebuilding from topk_ids

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:26:24 +00:00

f29b96de09 bug fixes

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:11:10 +00:00

a780bb5fde bug fix

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 09:09:39 +00:00

91338428d9 some optimizations

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 08:57:55 +00:00

fae418c3a3 final scatter

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 08:51:26 +00:00

f2cacfc2f2 fix the L2 path and the clamping math

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 08:28:52 +00:00

d22dae2df3 were getting close

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 08:02:05 +00:00

d493193d25 fix the god damn projections

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 07:45:37 +00:00

9810de7109 more debug

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 07:23:14 +00:00

1a37b66922 dang python

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 07:10:16 +00:00

7b3a853465 more debugging

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 06:40:40 +00:00

6b4b59c6a4 double check that weird line

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel

2026-05-15 06:38:23 +00:00

beacc31569 is paris in the top n?