DeepSeek模型用创新架构与训练方法，低成本实现顶尖大模型性能。
A Review of DeepSeek Models' Key Innovative Techniques
- 改进Transformer结构，引入多头潜在注意力与专家混合机制。
- 通过强化学习与迭代微调，实现媲美闭源模型的推理能力。
- 适合关注高效大模型研发、训练成本优化的研究者参考。
DeepSeek-V3 和 DeepSeek-R1 是面向通用任务和推理的领先开源大语言模型，性能可与 OpenAI、Anthropic 等公司最先进的闭源模型比肩，同时训练成本仅为后者的极小部分。本文系统回顾了推动这些模型高效性与强大表现的核心技术，包括对 Transformer 架构的优化、多头潜在注意力（Multi-Head Latent Attention）与专家混合（Mixture of Experts）等创新，以及多标记预测、算法-框架-硬件协同设计、组相对策略优化（Group Relative Policy Optimization）算法，还有仅使用强化学习的后训练及监督微调与强化学习交替的迭代训练策略。此外，本文还指出当前未解问题，并展望该快速发展的领域的潜在研究方向。
原文摘要 · Abstract (English)
DeepSeek-V3 and DeepSeek-R1 are leading open-source Large Language Models (LLMs) for general-purpose tasks and reasoning, achieving performance comparable to state-of-the-art closed-source models from companies like OpenAI and Anthropic -- while requiring only a fraction of their training costs. Understanding the key innovative techniques behind DeepSeek's success is crucial for advancing LLM research. In this paper, we review the core techniques driving the remarkable effectiveness and efficiency of these models, including refinements to the transformer architecture, innovations such as Multi-Head Latent Attention and Mixture of Experts, Multi-Token Prediction, the co-design of algorithms, frameworks, and hardware, the Group Relative Policy Optimization algorithm, post-training with pure reinforcement learning and iterative training alternating between supervised fine-tuning and reinforcement learning. Additionally, we identify several open questions and highlight potential research opportunities in this rapidly advancing field.
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。