研究人员正在开发新方法来提高大语言模型的效率和推理能力。一种名为ReHoPER的方法会在提供最终答案之前，沿着多条路径生成并回答中间问题，在组合推理任务上显示出优势。另一种名为Settle的方法，训练模型判断其推理何时稳定，在MATH-500数据集上将token数量减少了40%，准确率损失极小。第三种技术使用自监督置信度训练，鼓励模型预测其对答案的置信度，在不明确优化缩短推理的情况下，在各种模型和基准测试中实现了高达25%的效率提升。 AI
影响 这些技术可能带来更高效、更强大的大语言模型，降低计算成本并提高在复杂推理任务上的性能。
排序理由 多篇研究论文介绍了改进大语言模型推理和效率的新颖方法。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2610.00940v1 Announce Type: cross Abstract: We propose ReHoPER, an inference-only, zero-shot method that improves large language models' reasoning by generating and answering intermediate questions along multiple paths before the final answer. It iteratively plans a horizon…
arXiv:2609.38997v1 Announce Type: new Abstract: Reasoning models often continue generating after their answers have settled. Settle learns when to stop from answer stability in completed traces. It trains the existing end-of-reasoning token while keeping other predictions close t…
Reasoning models often continue generating after their answers have settled. Settle learns when to stop from answer stability in completed traces. It trains the existing end-of-reasoning token while keeping other predictions close to the base model, and requires only ordinary dec…
arXiv:2609.31619v1 Announce Type: new Abstract: Reasoning models often generate very long reasoning traces, making inference computationally expensive. Existing approaches typically improve efficiency either through inference-time early-stopping mechanisms or by explicitly encour…