Proprietary language models exhibit strong reasoning and coding capabilities, but their output distributions are inaccessible, making fine-grained distillati…
机构:中科院
来源:arXiv 2611.06542 | AI4Papers 论文推荐平台