控制LLM推理努力：多模式推理模型的开发方法

AI 摘要

本文探讨了如何控制大语言模型（LLM）的推理努力程度，使其具备低、中、高多种推理模式。文章回顾了推理模型的发展，如OpenAI的o1和DeepSeek-R1，并介绍了通过强化学习（RLVR）训练推理模型的方法，以及推理时计算扩展的概念。作者还提到了GPT-5.6系列模型提供多种推理努力设置，并计划深入讲解多努力模式模型的开发方法。

正文节选

Controlling Reasoning Effort in LLMs How LLMs Learn Low-, Medium-, and High-Effort Reasoning Modes It has been almost two years since OpenAI released o1, a model that popularized the idea of LLM-based reasoning models. DeepSeek-R1 followed about four months later, together with details of a reinforcement learning with verifiable rewards (RLVR) recipe to train such reasoning models. Last week, OpenAI released the GPT-5.6 model family. It comes in three sizes, each with roughly five or six reasoni

发布时间：2026-07-18 19:16

抓取时间：2026-08-02 00:27

来源机构：Sebastian Raschka