transformers
fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation
#45441

Merged

fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation #45441

ArthurZucker merged 2 commits into huggingface:main from casinca:dsv3-fix

fix(`DeepseekV3MoE`): correct expert masking with negative bias

7361e20f

Merge branch 'main' into dsv3-fix

5a50d3fa

vasqu commented on 2026-04-14

ArthurZucker approved these changes on 2026-04-22

ArthurZucker enabled auto-merge 195 days ago

ArthurZucker merged cb0adddc into main 195 days ago

casinca deleted the dsv3-fix branch 195 days ago

Reviewers

ArthurZucker

vasqu

Assignees

No one assigned

Labels

None yet

Milestone

No milestone