Proposes reasoning-prefix masking during VLM distillation to anchor student thinking on visual evidence and improve multimodal reasoning in smaller models.
Beyond next-token alignment: Distilling multimodal large language models via token interactions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Hide to See: Reasoning-prefix Masking for Visual-anchored Thinking in VLM Distillation
Proposes reasoning-prefix masking during VLM distillation to anchor student thinking on visual evidence and improve multimodal reasoning in smaller models.