References

本文为论文《Kimi K2：面向智能体能力的开源大模型》「中文全文翻译」板块的对应章节，与「论文原文」板块同一章节一一对应，层级保持一致。

  • [1] Jacob Austin et al.

“使用大语言模型进行程序合成（Program Synthesis with Large Language Models）”, 2021

arXiv: https://arxiv.org/abs/2108.07732

  • [2] Yushi Bai et al.

“LongBench v2：面向真实长上下文多任务上更深层次的理解与推理（LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks）”, 2025

arXiv: https://arxiv.org/abs/2412.15204

  • [3] Victor Barres et al.

“$\tau^{2}$-Bench：在双控制环境中评估对话智能体（$\tau^{2}$-Bench: Evaluating Conversational Agents in a Dual-Control Environment）”, 2025

arXiv: https://arxiv.org/abs/2506.07982

  • [4] Stella Biderman et al.

“可复现语言模型评估的实战经验（Lessons from the trenches on reproducible evaluation of language models）”

In arXiv preprint arXiv:2405.14782, 2024

  • [5] Federico Cassano et al.

“MultiPL-E：一种可扩展且支持多语言的方法来基准测试神经代码生成（MultiPL-E: A Scalable and Polyglot Approach to Benchmarking Neural Code Generation）”

In IEEE Transactions on Software Engineering 49.7, 2023, pp. 3675–3691

DOI: 10.1109/TSE.2023.3267446

  • [6] Chen Chen et al.

“ACEBench：谁在工具学习的赛点中胜出？（ACEBench: Who Wins the Match Point in Tool Learning?）”

In arXiv e-prints, 2025, pp. arXiv–2501

  • [7] Mark Chen et al.

“评估在代码上训练的大语言模型（Evaluating Large Language Models Trained on Code）”, 2021

arXiv:2107.03374 [cs.LG]

  • [8] Peter Clark et al.

“你认为自己已经解决了问答问题吗？试试 ARC，这个 AI2 推理挑战（Think you have solved question answering? try arc, the ai2 reasoning challenge）”

In arXiv preprint arXiv:1803.05457, 2018

  • [9] Karl Cobbe et al.

“训练验证器以解决数学应用题（Training Verifiers to Solve Math Word Problems）”, 2021

arXiv: https://arxiv.org/abs/2110.14168

  • [10] DeepSeek-AI

“DeepSeek-V3 技术报告（DeepSeek-V3 Technical Report）”, 2024

arXiv: https://arxiv.org/abs/2412.19437

  • [11] Mostafa Dehghani et al.

“将视觉 Transformer 扩展到 220 亿参数（Scaling vision transformers to 22 billion parameters）”

In International conference on machine learning, 2023, pp. 7480–7512

PMLR

  • [12] Guanting Dong et al.

“带执行反馈的自我博弈：提升大语言模型的指令遵循能力（Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models）”, 2024

arXiv: https://arxiv.org/abs/2406.13542

  • [13] Xinrun Du et al.

“SuperGPQA：在 285 个研究生学科上扩展 LLM 评估（Supergpqa: Scaling llm evaluation across 285 graduate disciplines）”

In arXiv preprint arXiv:2502.14739, 2025

  • [14] Dheeru Dua et al.

“DROP：一个需要对段落进行离散推理的阅读理解基准（DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs）”

In CoRR abs/1903.00161, 2019

arXiv: http://arxiv.org/abs/1903.00161

  • [15] Kazuki Fujii et al.

“重写预训练数据可提升 LLM 在数学和代码上的表现（Rewriting Pre-Training Data Boosts LLM Performance in Math and Code）”, 2025

arXiv: https://arxiv.org/abs/2505.02881

  • [16] Paul Gauthier

“Aider LLM 排行榜（Aider LLM Leaderboards）”

blog, https://aider.chat/docs/leaderboards/, 2025

  • [17] Aryo Gema et al.

“我们完成 MMLU 了吗？（Are we done with mmlu?）”

In arXiv preprint arXiv:2406.04127, 2024

  • [18] Alex Gu et al.

“CruxEval：一个用于代码推理、理解与执行的基准（Cruxeval: A benchmark for code reasoning, understanding and execution）”

In arXiv preprint arXiv:2401.03065, 2024

  • [19] Daya Guo et al.

“DeepSeek-R1：通过强化学习激励 LLM 的推理能力（Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning）”

In arXiv preprint arXiv:2501.12948, 2025

  • [20] Zhicheng Guo et al.

“StableToolBench：面向大语言模型工具学习的大规模稳定基准测试（StableToolBench: Towards Stable Large-Scale Benchmarking on Tool Learning of Large Language Models）”

In arXiv preprint arXiv:2403.07714, 2025

  • [21] Aaron Harlap et al.

“PipeDream：快速高效的数据流水线并行 DNN 训练（Pipedream: Fast and efficient pipeline parallel dnn training）”

In arXiv preprint arXiv:1806.03377, 2018

  • [22] Y He et al.

“Chinese SimpleQA：面向大语言模型的中文事实性评估（Chinese simpleqa: A chinese factuality evaluation for large language models, 2024a）”

In URL https://arxiv. org/abs/2411.07140

  • [23] Dan Hendrycks et al.

“测量大规模多任务语言理解（Measuring massive multitask language understanding）”

In arXiv preprint arXiv:2009.03300, 2020

  • [24] Dan Hendrycks et al.

“用 MATH 数据集衡量数学问题求解能力（Measuring Mathematical Problem Solving With the MATH Dataset）”, 2021

arXiv: https://arxiv.org/abs/2103.03874

  • [25] Shengding Hu et al.

“MiniCPM：通过可扩展训练策略揭示小语言模型的潜力（Minicpm: Unveiling the potential of small language models with scalable training strategies）”

In arXiv preprint arXiv:2404.06395, 2024

  • [26] Jiaxin Huang et al.

“大语言模型可以自我改进（Large language models can self-improve）”

In arXiv preprint arXiv:2210.11610, 2022

  • [27] Siming Huang et al.

“OpenCoder：顶级代码大语言模型的开源秘籍（OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models）”, 2025

arXiv: https://arxiv.org/abs/2411.04905

  • [28] Yanping Huang et al.

“GPipe：使用流水线并行高效训练巨型神经网络（Gpipe: Efficient training of giant neural networks using pipeline parallelism）”

In Advances in neural information processing systems 32, 2019

  • [29] Yuzhen Huang et al.

“C-Eval：一个面向基础模型的多层级、多学科中文评估套件（C-Eval: A Multi-Level Multi-Discipline Chinese Evaluation Suite for Foundation Models）”, 2023

arXiv: https://arxiv.org/abs/2305.08322

  • [30] Alon Jacovi et al.

“FACTS Grounding 排行榜：基准测试 LLM 将长文输入响应落到实处的的能力（The FACTS Grounding Leaderboard: Benchmarking LLMs’ Ability to Ground Responses to Long-Form Input）”, 2025

arXiv: https://arxiv.org/abs/2501.03200

  • [31] Naman Jain et al.

“LiveCodeBench：对代码大语言模型进行整体且无污染的评估（Livecodebench: Holistic and contamination free evaluation of large language models for code）”

In arXiv preprint arXiv:2403.07974, 2024

  • [32] Carlos Jimenez et al.

“SWE-bench：语言模型能否解决真实世界 GitHub 问题？（SWE-bench: Can Language Models Resolve Real-world Github Issues?）”

In The Twelfth International Conference on Learning Representations, 2024

URL: https://openreview.net/forum?id=VTF8yNQM66

  • [33] Keller Jordan et al.

“Muon：一种用于神经网络隐藏层的优化器（Muon: An optimizer for hidden layers in neural networks）”, 2024

URL: https://kellerjordan.github.io/posts/muon/

  • [34] Mandar Joshi et al.

“TriviaQA：一个大规模远程监督的阅读理解挑战数据集（TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension）”, 2017

arXiv: https://arxiv.org/abs/1705.03551

  • [35] Kimi Team

“Kimi K1.5：使用 LLM 扩展强化学习（Kimi k1. 5: Scaling reinforcement learning with llms）”

In arXiv preprint arXiv:2501.12599, 2025

  • [36] Diederik. Kingma and Jimmy Ba

“Adam：一种随机优化方法（Adam: A Method for Stochastic Optimization）”

In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015

URL: http://arxiv.org/abs/1412.6980

  • [37] Satyapriya Krishna et al.

“Fact, Fetch, and Reason：检索增强生成的统一评估（Fact, Fetch, and Reason: A Unified Evaluation of Retrieval-Augmented Generation）”, 2025

arXiv: https://arxiv.org/abs/2409.12941

  • [38] Joel Lamy-Poirier

“广度优先流水线并行（Breadth-first pipeline parallelism）”

In Proceedings of Machine Learning and Systems 5, 2023, pp. 48–67

  • [39] Dmitry Lepikhin et al.

“GShard：利用条件计算与自动分片扩展巨型模型（Gshard: Scaling giant models with conditional computation and automatic sharding）”

In arXiv preprint arXiv:2006.16668, 2020

  • [40] Haonan Li et al.

“CMMLU：在中文中测量大规模多任务语言理解（CMMLU: Measuring massive multitask language understanding in Chinese）”, 2024

arXiv: https://arxiv.org/abs/2306.09212

  • [41] Jia Li et al.

“NuminaMath：AI4Maths 中最大的公开数据集，包含 86 万组竞赛数学问题与解答（Numinamath: The largest public dataset in ai4maths with 860k pairs of competition math problems and solutions）”

In Hugging Face repository 13.9, 2024, pp. 9

  • [42] Tianle Li et al.

“从众包数据到高质量基准：Arena-Hard 与 BenchBuilder 流水线（From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline）”

In arXiv preprint arXiv:2406.11939, 2024

  • [43] Bill Lin et al.

“ZebraLogic：关于 LLM 逻辑推理的扩展极限（ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning）”, 2025

arXiv: https://arxiv.org/abs/2502.01100

  • [44] Aixin Liu et al.

“DeepSeek-V2：一个强大、经济且高效的混合专家语言模型（Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model）”

In arXiv preprint arXiv:2405.04434, 2024

  • [45] Jiawei Liu et al.

“你用 ChatGPT 生成的代码真的正确吗？对大语言模型代码生成的严格评估（Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation）”

In Advances in Neural Information Processing Systems 36, 2023, pp. 21558–21572

  • [46] Jingyuan Liu et al.

“Muon 可用于 LLM 训练扩展（Muon is scalable for LLM training）”

In arXiv preprint arXiv:2502.16982, 2025

  • [47] Ziming Liu et al.

“Hanayo：利用波浪式流水线并行提升大模型训练效率（Hanayo: Harnessing Wave-like Pipeline Parallelism for Enhanced Large Model Training Efficiency）”

In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC ’23

ACM, 2023, pp. 1–13

DOI: 10.1145/3581784.3607073

  • [48] Ilya Loshchilov and Frank Hutter

“解耦权重衰减正则化（Decoupled Weight Decay Regularization）”

In International Conference on Learning Representations, 2019

URL: https://openreview.net/forum?id=Bkg6RiCqY7

  • [49] Jan Ludziejewski et al.

“OpenAI Gym”, 2025

arXiv: https://arxiv.org/abs/2502.05172

  • [50] Samuel Miserendino et al.

“SWE-Lancer：前沿 LLM 能否从真实世界自由职业软件工程中赚取 100 万美元？（SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?）”

In arXiv preprint arXiv:2502.12115, 2025

  • [51] Arindam Mitra et al.

“AgentInstruct：迈向带智能体流的生成式教学（Agentinstruct: Toward generative teaching with agentic flows）”

In arXiv preprint arXiv:2407.03502, 2024

  • [52] Ivan Moshkov et al.

“AIMO-2 获胜方案：使用 OpenMathReasoning 数据集构建最先进的数学推理模型（Aimo-2 winning solution: Building state-of-the-art mathematical reasoning models with openmathreasoning dataset）”

In arXiv preprint arXiv:2504.16891, 2025

  • [53] Deepak Narayanan et al.

“使用 Megatron-LM 在 GPU 集群上高效进行大规模语言模型训练（Efficient large-scale language model training on gpu clusters using megatron-lm）”

In Proceedings of the international conference for high performance computing, networking, storage and analysis, 2021, pp. 1–15

  • [54] Long Ouyang et al.

“训练语言模型以遵循人类反馈的指令（Training language models to follow instructions with human feedback）”

In Advances in neural information processing systems 35, 2022, pp. 27730–27744

  • [55] Bowen Peng et al.

“YaRN：大语言模型上下文窗口的高效扩展（Yarn: Efficient context window extension of large language models）”

In arXiv preprint arXiv:2309.00071, 2023

  • [56] Long Phan et al.

“人类最后的考试（Humanity’s Last Exam）”, 2025

arXiv: https://arxiv.org/abs/2501.14249

  • [57] Penghui Qi et al.

“零气泡流水线并行（Zero bubble pipeline parallelism）”

In arXiv preprint arXiv:2401.10241, 2023

  • [58] Yujia Qin et al.

“ToolLLM：助力大语言模型掌握 16000+ 真实世界 API（Toolllm: Facilitating large language models to master 16000+ real-world apis）”

In arXiv preprint arXiv:2307.16789, 2023

  • [59] Qwen et al.

“Qwen2.5 技术报告（Qwen2.5 Technical Report）”, 2025

arXiv: https://arxiv.org/abs/2412.15115

  • [60] Samyam Rajbhandari et al.

“ZeRO：面向万亿参数模型训练的内存优化（Zero: Memory optimizations toward training trillion parameter models）”

In SC20: International Conference for High Performance Computing, Networking, Storage and Analysis, 2020, pp. 1–16

IEEE

  • [61] David Rein et al.

“GPQA：一个研究生级别的、防 Google 的问答基准（Gpqa: A graduate-level google-proof q&a benchmark）”

In First Conference on Language Modeling, 2024

  • [62] Keisuke Sakaguchi et al.

“WinoGrande：大规模对抗性 Winograd 模式挑战（Winogrande: An adversarial winograd schema challenge at scale）”

In Communications of the ACM 64.9

ACM New York, NY, USA, 2021, pp. 99–106

  • [63] David Silver and Richard Sutton

“欢迎来到经验的时代（Welcome to the era of experience）”

In Google AI 1, 2025

  • [64] Ved Sirdeshmukh et al.

“MultiChallenge：一个对前沿 LLM 而言具有挑战性的真实多轮对话评估基准（MultiChallenge: A Realistic Multi-Turn Conversation Evaluation Benchmark Challenging to Frontier LLMs）”, 2025

arXiv: https://arxiv.org/abs/2501.17399

  • [65] Giulio Starace et al.

“PaperBench：评估 AI 复现 AI 研究的能力（PaperBench: Evaluating AI’s Ability to Replicate AI Research）”

In arXiv preprint arXiv:2504.01848, 2025

  • [66] Hao Sun et al.

“ZeroSearch：无需实际搜索即可激励 LLM 的搜索能力（ZeroSearch: Incentivize the Search Capability of LLMs without Searching）”, 2025

arXiv: https://arxiv.org/abs/2505.04588

  • [67] Mirac Suzgun et al.

“具有挑战性的 BIG-Bench 任务以及思维链能否解决它们（Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them）”, 2022

arXiv: https://arxiv.org/abs/2210.09261

  • [68] Manveer Tamber et al.

“使用演进式排行榜基准测试 RAG 中 LLM 的忠实性（Benchmarking LLM Faithfulness in RAG with Evolving Leaderboards）”

In arXiv preprint arXiv:2505.04847, 2025

  • [69] Gemma Team et al.

“Gemma 2：在实际规模下改进开放语言模型（Gemma 2: Improving open language models at a practical size）”

In arXiv preprint arXiv:2408.00118, 2024

  • [70] LlaMA Team

“Llama 4 系列：原生多模态 AI 创新新纪元的开端 —— ai.meta.com” [Accessed 15-07-2025], https://ai.meta.com/blog/llama-4-multimodal-intelligence/

  • [71] The-Bench Team

“Terminal-Bench：终端环境中的 AI 智能体基准（Terminal-Bench: A Benchmark for AI Agents in Terminal Environments）”, 2025

URL: https://github.com/laude-institute/terminal-bench

  • [72] Ashish Vaswani et al.

“Attention is All you Need（注意力即你所需要的一切）”

In Advances in Neural Information Processing Systems 30

Curran Associates, Inc., 2017

URL: https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf

  • [73] Vectara

“幻觉评估模型（修订版 7437011）（Hallucination Evaluation Model (Revision 7437011)）”

Hugging Face, 2024

URL: https://huggingface.co/vectara/hallucination_evaluation_model

  • [74] Joshua Vendrow et al.

“大语言模型基准测试是否测试了可靠性？（Do large language model benchmarks test reliability?）”

In arXiv preprint arXiv:2502.03461, 2025

  • [75] Yizhong Wang et al.

“Self-Instruct：用自生成指令对齐语言模型（Self-instruct: Aligning language models with self-generated instructions）”

In arXiv preprint arXiv:2212.10560, 2022

  • [76] Yubo Wang et al.

“MMLU-Pro：一个更鲁棒、更具挑战性的多任务语言理解基准（MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark）”, 2024

arXiv: https://arxiv.org/abs/2406.01574

  • [77] Zhexu Wang et al.

“OJBench：一个面向大语言模型的竞赛级代码基准（OJBench: A Competition Level Code Benchmark For Large Language Models）”, 2025

arXiv: https://arxiv.org/abs/2506.16395

  • [78] Jason Wei et al.

“测量大语言模型的短文本事实性（Measuring short-form factuality in large language models）”

In arXiv preprint arXiv:2411.04368, 2024

  • [79] Tianwen Wei et al.

“CMATH：你的大语言模型能通过中国小学数学测试吗？（CMATH: Can Your Language Model Pass Chinese Elementary School Math Test?）”, 2023

arXiv: https://arxiv.org/abs/2306.16636

  • [80] Colin White et al.

“LiveBench：一个具有挑战性、无污染的 LLM 基准（LiveBench: A Challenging, Contamination-Free LLM Benchmark）”

In The Thirteenth International Conference on Learning Representations, 2025

  • [81] Mitchell Wortsman et al.

“用于大规模 Transformer 训练不稳定性的小尺度代理（Small-scale proxies for large-scale transformer training instabilities, 2023）”

In URL https://arxiv. org/abs/2309.14322

  • [82] Can Xu et al.

“WizardLM：赋能大型预训练语言模型以遵循复杂指令（WizardLM: Empowering large pre-trained language models to follow complex instructions）”, 2025

arXiv: https://arxiv.org/abs/2304.12244

  • [83] Zhangchen Xu et al.

“KodCode：一个多样、具有挑战性且可验证的编码合成数据集（KodCode: A Diverse, Challenging, and Verifiable Synthetic Dataset for Coding）”, 2025

arXiv: https://arxiv.org/abs/2503.02951

  • [84] John Yang et al.

“SWE-Smith：为软件工程智能体扩展数据（SWE-smith: Scaling Data for Software Engineering Agents）”, 2025

arXiv: https://arxiv.org/abs/2504.21798

  • [85] Shunyu Yao et al.

“$\tau$-bench：真实世界领域中工具-智能体-用户交互的基准（$\tau$-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains）”

In arXiv preprint arXiv:2406.12045, 2024

  • [86] Daoguang Zan et al.

“Multi-SWE-bench：一个用于问题修复的多语言基准（Multi-swe-bench: A multilingual benchmark for issue resolving）”

In arXiv preprint arXiv:2504.02605, 2025

  • [87] Eric Zelikman et al.

“STAR：用推理引导推理（Star: Bootstrapping reasoning with reasoning）”

In Advances in Neural Information Processing Systems 35, 2022, pp. 15476–15488

  • [88] Rowan Zellers et al.

“HellaSwag：机器真的能补全你的句子吗？（Hellaswag: Can a machine really finish your sentence?）”

In arXiv preprint arXiv:1905.07830, 2019

  • [89] Wanjun Zhong et al.

“AGIEval：一个以人为中心的基础模型评估基准（Agieval: A human-centric benchmark for evaluating foundation models）”

In arXiv preprint arXiv:2304.06364, 2023

  • [90] Jeffrey Zhou et al.

“面向大语言模型的指令遵循评估（Instruction-Following Evaluation for Large Language Models）”

In ArXiv abs/2311.07911, 2023

URL: https://arxiv.org/abs/2311.07911

  • [91] Qin Zhu et al.

“AutoLogi：自动化生成逻辑谜题以评估大语言模型的推理能力（AutoLogi: Automated Generation of Logic Puzzles for Evaluating Reasoning Abilities of Large Language Models）”, 2025

arXiv: https://arxiv.org/abs/2502.16906