模型14:08无训练循环Transformer提升冻结AI模型推理能力这篇论文教会你一种骚操作：不重新训练，就能让现成模型在推理时多思考几轮，MMLU-Pro和GPQA分数都涨了，值得看看。#Looped Transformers#MMLU-Pro#GPQA#推理模型1 个信源在谈事件专题官方账号AlphaSignal@AlphaSignalAI2 个信源在谈原文稍后读已读值得跟进有用关注 Looped Transformers
