Qwen2.5-1M Technical Report
打开互动全文版（逐段中英对照 + 图/公式 + 论文问答）→我们推出了 Qwen2.5-1M 系列模型，其上下文长度扩展至 100 万 token。与之前的 128K 版本相比，Qwen2.5-1M 系列通过长上下文预训练和后训练显著增强了长上下文能力。我们采用了长数据合成、渐进式预训练和多阶段监督微调等关键技术，在降低训练成本的同时有效提升长上下文性能。为了向更广泛的用户群体推广长上下文模型，我们提出并开源了推理框架。该框架包含一种长度外推方法，无需额外训练即可将模型上下文长度扩展至少四倍，甚至更多。为降低推理成本，我们针对部署场景实现了稀疏注意力和分块预填充优化，并通过稀疏度精炼方法提高精度。此外，我们详细描述了推理引擎中的优化，包括内核优化、流水线并行和调度优化，这些显著提升了整体推理性能。借助我们的推理框架，Qwen2.5-1M 模型在 100 万 token 上下文场景下实现了 3 到 7 倍的预填充加速。该框架为使用开源模型开发需要长上下文处理的应用提供了高效而强大的解决方案。Qwen2.5-1M 系列目前包括开源模型 Qwen2.5-7B-Instruct-1M 和 Qwen2.5-14B-Instruct-1M，以及通过 API 访问的模型 Qwen2.5-Turbo。评估表明，Qwen2.5-1M 模型在长上下文任务上取得了显著改进，同时未损害短上下文场景的性能。具体来说，Qwen2.5-14B-Instruct-1M 模型在长上下文任务上显著优于 GPT-4o-mini，并支持八倍于后者的上下文长度。
We introduce Qwen2.5-1M, a series of models that extend the context length to 1 million tokens. Compared to the previous 128K version, the Qwen2.5-1M series have significantly enhanced long-context capabilities through long-context pre-training and post-training. Key techniques such as long data synthesis, progressive pre-training, and multi-stage supervised fine-tuning are employed to effectively enhance long-context performance while reducing training costs. To promote the use of long-context models among a broader user base, we present and open-source our inference framework. This framework includes a length extrapolation method that can expand the model context lengths by at least four times, or even more, without additional training. To reduce inference costs, we implement a sparse attention method along with chunked prefill optimization for deployment scenarios and a sparsity refinement method to improve precision. Additionally, we detail our optimizations in the inference engine, including kernel optimization, pipeline parallelism, and scheduling optimization, which significantly enhance overall inference performance. By leveraging our inference framework, the Qwen2.5-1M models achieve a remarkable 3x to 7x prefill speedup in scenarios with 1 million tokens of context. This framework provides an efficient and powerful solution for developing applications that require long-context processing using open-source models. The Qwen2.5-1M series currently includes the open-source models Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M, as well as the API-accessed model Qwen2.5-Turbo. Evaluations show that Qwen2.5-1M models have been greatly improved in long-context tasks without compromising performance in short-context scenarios. Specifically, the Qwen2.5-14B-Instruct-1M model significantly outperforms GPT-4o-mini in long-context tasks and supports contexts eight times longer.