biondizzle

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

2026-05-16 01:15:50 +00:00

4a624879ca docs: update DEBUG_LOG — input_scale red herring, current state, next steps

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

2026-05-16 00:12:44 +00:00

79b9becf9c revert: don't use checkpoint input_scale for activation normalization

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

2026-05-15 23:57:10 +00:00

a7eae10ef4 fix: use checkpoint input_scale for activation quantization

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

2026-05-15 23:52:23 +00:00

af50e98fe9 test: B layout test with N=128 K=256

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

2026-05-15 23:52:02 +00:00

efd7a2c56d test: B matrix weight layout verification via one-hot A

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

2026-05-15 23:50:40 +00:00

bb5a1ba4c8 cleanup: remove unused slot_token from nvfp4_moe_l2

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

2026-05-15 23:38:35 +00:00

887360281e docs: major update — SF remap verified correct, BF16 ref is the red herring

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

2026-05-15 23:36:10 +00:00

eb26d291cb test: uniform FP4 + uniform SF sanity check

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

2026-05-15 23:35:09 +00:00

1f09b51168 test: check SF signed vs unsigned interpretation

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

2026-05-15 23:02:33 +00:00

4f857d5f99 docs: major DEBUG_LOG update — forward mapping, verifier, full debug timeline

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

2026-05-15 22:59:52 +00:00

aa209ddd21 debug: add SF remap roundtrip verifier

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

2026-05-15 22:58:52 +00:00

6626b75a2f fix: use filter_zeros for SF allocation + no-branch forward mapping

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

2026-05-15 22:53:59 +00:00

6fc8fa61e0 fix: use flat logical coordinate layout_sf(make_coord(mn, k_elem, 0))

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

2026-05-15 22:31:06 +00:00

a48717ccf5 fix: remove duplicate dst_idx declaration

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

2026-05-15 22:11:15 +00:00

5ff1b9e401 fix: use hierarchical coordinates for layout_sf forward mapping

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

2026-05-15 22:09:57 +00:00

3b4a7b591f test: verify forward mapping with prepack vs live SFB

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

2026-05-15 21:55:03 +00:00

a1fd4d6233 revert: back to layout_sf(make_coord(...)) — crd2idx was unnecessary

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

2026-05-15 21:54:12 +00:00

ea678ece64 fix: remove duplicate template declaration

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

2026-05-15 21:52:05 +00:00

59dad8e2fb fix: use crd2idx instead of layout operator() for SF forward mapping

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

2026-05-15 21:27:30 +00:00

a09d8e477e fix: remove static_assert in constexpr else (build fix)