Add return_attention support to Generator#2010

Open

QuentinFuxa wants to merge 3 commits into

OpenNMT:masterfrom

QuentinFuxa:feature/generator-return-attention

Commits on Feb 7, 2026

Add return_attention support to Generator
QuentinFuxa
committed

Commits on Feb 20, 2026

Fix self-attention for decoder-only models and add set_alignment_heads API
QuentinFuxa
committed

Commits on May 27, 2026

Merge branch 'master' into feature/generator-return-attention
QuentinFuxa
authored