On-policy self-distillation improves a model using guidance from the model itself, but it still requires a teacher that provides more useful predictions than…
机构:阿里
来源:arXiv 2610.18162 | AI4Papers 论文推荐平台