Adaptive distillation teaches smaller AI models more reliably

Original: On-Policy Distillation for Vision-Language Model Adaptation, an Effective Paradigm on Low-Quality Multimodal Data

Source: arxiv.org ↗