Jul 2026
OPLD: On-Policy Latent Distillation for Multimodal Reasoning
OPLD (On-Policy Latent Distillation), a simple framework that transfers the reasoning capability induced by privileged multimodal CoT into latent reasoning representations, is proposed and extensive experiments demonstrate that OPLD consistently outperforms existing latent reasoning methods and achieves state-of-the-ar...
Shoutai Zhu, Tianyang Xu, Bingchuan Sun et al.
· arXiv.org · 0 citations