ToiTenBao's picture
Upload hallucination folder
a2ffd07 verified
Raw
History Blame Contribute Delete
507 Bytes
"""
DualEdit baseline for multimodal knowledge editing.
Adapted from: https://github.com/zhiyiscs/DualEdit
Paper: "DualEdit: Dual Editing for Knowledge Updating in Vision-Language Models" (COLM 2025)
This adaptation:
- Reuses the cross-attention adapter architecture from the original DualEdit
- Adapts training from per-sample edits to distributional (batched) editing
- Integrates with our run_baselines.py / validate.py pipeline
"""
from .dualedit_main import apply_dualedit_to_multimodal_model