Hermes 4 融合多轮推理与指令遵循，提升复杂任务处理能力。
Hermes 4 Technical Report
- 采用混合架构实现结构化推理与广泛指令理解
- 在数学、编程等五类基准上表现优于同类模型
- 适合需要强推理与精准执行的AI研究者使用
我们提出Hermes 4，一类结合结构化多轮推理与广泛指令遵循能力的混合推理模型。文中详述了数据整理、合成、训练与评估中遇到的挑战，并介绍了规模化解决方案。我们在数学推理、编程、知识、理解与对齐等多个基准上进行综合评估，报告了定量性能与定性行为分析结果。为支持开放研究，所有模型权重已公开发布于 https://huggingface.co/collections/NousResearch/hermes-4-collection-68a731bfd452e20816725728。
原文摘要 · Abstract (English)
We present Hermes 4, a family of hybrid reasoning models that combine structured, multi-turn reasoning with broad instruction-following ability. We describe the challenges encountered during data curation, synthesis, training, and evaluation, and outline the solutions employed to address these challenges at scale. We comprehensively evaluate across mathematical reasoning, coding, knowledge, comprehension, and alignment benchmarks, and we report both quantitative performance and qualitative behavioral analysis. To support open research, all model weights are published publicly at https://huggingface.co/collections/NousResearch/hermes-4-collection-68a731bfd452e20816725728
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。