MiMo: Unlocking the Reasoning Potential of Language Model – From Pretraining to Posttraining

来源摘要

We present MiMo-7B, a large language model born for reasoning tasks, with optimization across both pre-training and post-training stages. During pre-training, we enhance the data preprocessing pipeline and employ a three-stage data mixing strategy to strengthen the base model’s r…

摘要由机器生成（来自来源站点），可能存在偏差； 本站不转载全文，请以原文为准。