跳到正文
返回论文列表
cs.AI提交于 已译

Qwen3 技术报告

Qwen3 Technical Report

Yang, An · A. Li · Baosong Yang · Beichen Zhang · Binyuan Hui · Zheng, Bo · et al.

中文摘要

Qwen3 是 Qwen 模型系列的最新版本,包含一系列大型语言模型(LLM),旨在提升性能、效率和多语言能力。该系列涵盖稠密(dense)和专家混合(MoE)架构,参数规模从 0.6 到 2350 亿。关键创新是将思考模式(用于复杂多步推理)和非思考模式(用于快速上下文驱动响应)集成到统一框架中,无需在聊天优化模型(如 GPT-4o)和专用推理模型(如 QwQ-32B)之间切换,并支持基于用户查询或聊天模板的动态模式切换。同时,Qwen3 引入思考预算机制,允许用户在推理时自适应分配计算资源,根据任务复杂度平衡延迟和性能。此外,利用旗舰模型的知识,显著减少构建较小规模模型所需的计算资源,同时确保其性能极具竞争力。实证评估表明,Qwen3 在代码生成、数学推理、智能体任务等多个基准上达到最先进结果,可与更大 MoE 模型和专有模型竞争。与 predecessor Qwen2.5 相比,Qwen3 将多语言支持从 29 种扩展到 119 种语言和方言,通过改进的跨语言理解和生成能力增强全球可访问性。为促进可复现性和社区驱动的研究与开发,所有 Qwen3 模型均在 Apache 2.0 下公开可用。

关键要点

  1. 01问题:需在聊天优化和专用推理模型间切换,且多语言支持有限。
  2. 02方法:统一思考与非思考模式,引入思考预算机制,涵盖稠密和 MoE 架构。
  3. 03结果:在代码生成、数学推理、智能体任务等基准上达到最先进,多语言支持扩至 119 种。
  4. 04局限:未明确提及。

解读

尚无解读。

原始英文摘要

In this work, we present Qwen3, the latest version of the Qwen model family. Qwen3 comprises a series of large language models (LLMs) designed to advance performance, efficiency, and multilingual capabilities. The Qwen3 series includes models of both dense and Mixture-of-Expert (MoE) architectures, with parameter scales ranging from 0.6 to 235 billion. A key innovation in Qwen3 is the integration of thinking mode (for complex, multi-step reasoning) and non-thinking mode (for rapid, context-driven responses) into a unified framework. This eliminates the need to switch between different models--such as chat-optimized models (e.g., GPT-4o) and dedicated reasoning models (e.g., QwQ-32B)--and enables dynamic mode switching based on user queries or chat templates. Meanwhile, Qwen3 introduces a thinking budget mechanism, allowing users to allocate computational resources adaptively during inference, thereby balancing latency and performance based on task complexity. Moreover, by leveraging the knowledge from the flagship models, we significantly reduce the computational resources required to build smaller-scale models, while ensuring their highly competitive performance. Empirical evaluations demonstrate that Qwen3 achieves state-of-the-art results across diverse benchmarks, including tasks in code generation, mathematical reasoning, agent tasks, etc., competitive against larger MoE models and proprietary models. Compared to its predecessor Qwen2.5, Qwen3 expands multilingual support from 29 to 119 languages and dialects, enhancing global accessibility through improved cross-lingual understanding and generation capabilities. To facilitate reproducibility and community-driven research and development, all Qwen3 models are publicly accessible under Apache 2.0.

同方向论文 · cs.AI

查看全部 →

反向发现 · 这篇论文相关的工具与提示词

去交叉发现 →

关联工具 · 1

量子位 QbitAI87

阿里发布Qwen4-Max基座模型,智能体与编程能力进阶

阿里巴巴11月16日发布新一代基座大模型Qwen4-Max,总参数2.4T、仅激活71B,具备原生视觉能力并支持百万上下文。该模型在CodeArena前端编程盲测中位列全球第6,在Agents' Last Exam取得46.7分,较上代提升18.8分。Qwen4-Max已上线千问AI平台提供API服务,缓存命中每百万Tokens低至0.5元,单任务平均费用较上代降低58%,中小尺寸模型将陆续开源。

推荐理由:读者可了解 Qwen4-Max 的架构取舍、编程与智能体评测位次,以及它在制造、科研场景中的落地案例。