transformers
Fix the misalignment between the l2norm in GDN of Qwen3-Next and the implementation in the FLA library.
#40842

Merged

Fix the misalignment between the l2norm in GDN of Qwen3-Next and the implementation in the FLA library. #40842

Cyrilvallez merged 7 commits into huggingface:main from bozheng-hit:qwen3_next_fix_torch_l2norm

align torch implementation of gdn with fla.

cf9bf796

fix fla import.

1af3e404

fix

e0c4db9e

remove unused attr

7cac277e

fixes

ab9f5380

Merge branch 'huggingface:main' into qwen3_next_fix_torch_l2norm

f536c93c

strictly align l2norm in Qwen3-Next with FLA implementation.

a0bf6f2f

Cyrilvallez approved these changes on 2025-09-12

Cyrilvallez merged 98a80781 into main 94 days ago

Reviewers

Cyrilvallez

Assignees

No one assigned

Labels

None yet

Milestone

No milestone