NVFP4 on the fine-grained machinery #49779
IlyasMoutawwakil
wants to merge
18 commits into
fg-split/6-mxfp4
from
fg-split/7-nvfp4
IlyasMoutawwakil
changed the title
Move NVFP4Config onto the fine-grained machinery, with the producers'…
fb79484a
Document NVFP4 on the fine-grained machinery and ModelOpt mixed-preci…
43ddf8d7
Delete the frozen NVFP4 integration
d83fbbf2
Lend NVFP4's activation format like its other loading attributes
164c37e5
Give the DeepGEMM refusal test's experts two K bands so MX scales swi…
d1b7081c
Skip the NVFP4 activation check when a plan load dequantizes
1f728460
Explain why NVFP4 inverts its globals and checks weight-only loads
4477b933
Test that DeepGEMM grouped experts refuse the Triton layout
1b7d4f44
Update the NVFP4 plan tests for the generic host
e7f30fc2
Fix the plan tests the host rebase merged twice
b6defe6c
Read the NVFP4 producer layouts through get_weight_conversions
cb8ecaf3
Name the embedding tables of the NVFP4 tests with targeted plan entries
f62e0ec7
Keep the old NVFP4 integration importable for a deprecation cycle
7bfee090
Let a post-expert norm declare its fused kernel form
883435d2
Give the table test a model that holds the table
1a1bfd5a
Leave NVFP4's compressed-tensors and ModelOpt export support to the p…
bc73e938
Keep the saved-FP8 DeepSeek-V4 check in the plan test that stays
16910d45
Restack on the simpler plan resolution and merging
9ee98c88
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub