Unlock DeepSeek 是一个面向广大 AI 研究者和学习者的开源学习项目，致力于对 DeepSeek 系列论文进行系统性解读与动手复现。项目涵盖 DeepSeek 在通用大语言模型、数学推理、代码生成、多模态、推理模型（如 DeepSeek-R1）以及 MoE 架构、训练基础设施等方向的创新成果，旨在将 DeepSeek 在 AGI 实践之路上的前沿技术拆解为可理解、可复现的学习内容。
核心特色：
- 论文精读：从 DeepSeek-LLM 到 DeepSeek-V3.2、DeepSeek-Math、DeepSeek-Coder、DeepSeek-VL 等全系列技术报告的系统解读
- 动手实践：提供从零开始的复现教程与代码，支持中文社区本地化复现（如 DeepSeek-R1 等）
- 技术拆解：聚焦 MoE 架构、模型推理与强化学习、数据与 Infra 等关键要素
- 横向对比：引入 Kimi, GLM, MiniMax 等同类工作比较，呈现 AGI 路径上的不同技术路线
目标读者：具备大语言模型基础知识和大学数学水平的初学者、希望深入理解深度推理的学习者，以及希望将推理模型应用于实际工作的从业者。
项目当前处于
| 章节 | 摘要 | 状态 |
|---|---|---|
| 引言 | 项目概述以及学习建议 | ✅ |
| 通用大语言模型 | ||
| DeepSeek-LLM | Scaling Open-Source Language Models with Longtermism | 🚧 |
| DeepSeekMoE | Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models | 🚧 |
| DeepSeek-V2 | A Strong, Economical, and Efficient Mixture-of-Experts Language Model | 🚧 |
| DeepSeek-V3 | DeepSeek-V3 Technical Report | 🚧 |
| DeepSeek-V3.2 | Pushing the Frontier of Open Large Language Models | 🚧 |
| DeepSeek-V3.2-Exp | Boosting Long-Context Efficiency with DeepSeek Sparse Attention | 🚧 |
| 数学 | ||
| DeepSeek-Math | Pushing the Limits of Mathematical Reasoning in Open Language Models | 🚧 |
| DeepSeek-Math-V2 | Towards Self-Verifiable Mathematical Reasoning | 🚧 |
| DeepSeek-Prover | Advancing Theorem Proving in LLMs through Large-Scale Synthetic Data | 🚧 |
| DeepSeek-Prover-V1.5 | Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search | 🚧 |
| DeepSeek-Prover-V2 | Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition | 🚧 |
| 代码 | ||
| DeepSeek-Coder | Let the Code Write Itself | 🚧 |
| DeepSeek-Coder-V2 | Breaking the Barrier of Closed-Source Models in Code Intelligence | 🚧 |
| 多模态 | ||
| DeepSeek-OCR | Contexts Optical Compression | 🚧 |
| DeepSeek-OCR-2 | Visual Causal Flow | 🚧 |
| DeepSeek-VL | Towards Real-World Vision-Language Understanding | 🚧 |
| DeepSeek-VL2 | Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding | 🚧 |
| Janus | Decoupling visual encoding for unified multimodal understanding and generation | 🚧 |
| JanusFlow | Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation | 🚧 |
| Janus-Pro | Unified Multimodal Understanding and Generation with Data and Model Scaling | 🚧 |
| 推理模型/强化学习 | ||
| DeepSeek-R1 | Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | ✅ |
| 模型架构与训练方法 | ||
| Engram | Conditional Memory via Scalable Lookup | 🚧 |
| mHC | Manifold-Constrained Hyper-Connections | 🚧 |
| FlashMLA | Efficient Multi-head Latent Attention Kernels | 🚧 |
| LPLB | Linear-Programming-Based Load Balancer | 🚧 |
| ESFT | Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models | 🚧 |
| INFRA | ||
| DualPipe | 🚧 | |
| 3FS | 🚧 | |
| DeepEP | 🚧 | |
| DeepGEMM | 🚧 | |
| 附录 | ||
| 附录内容 | ✅ |
感谢以下项目的核心贡献者，排名不分先后
- 骆秀韬 - 项目负责人 （似然实验室）
- 姜舒凡 - 项目负责人 （Datawhale 成员）
- 邓恺俊 （深圳大学）
- 陈嘉诺 （广州大学）
- 林景豪 （智谱）
本项目受益于 DeepSeek-R1, Open-R1, transformers, pytorch, flash-attn, trl, mini-deepseek-r1, TinyZero, modelscope, vllm 感谢以上开源项目的杰出工作！
项目采用知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议进行许可。