tracked author
Damai Dai
Math-Shepherd coauthor and DeepSeek MoE / LLM researcher. OpenReview records DeepSeek-AI researcher, Peking University PhD and undergraduate background, and Zhifang Sui as PhD advisor. No high-confidence personal X account found in this pass.
6 archived notes X: not-found
Related Notes
按论文归档时间排序，展示该作者在本站已经出现的材料。
归档 更新
DeepSeek V4.1 Flash: Pushing the Limits of KV Cache Compression
DeepSeek V4.1 Flash 将因果编码器与解码器分工、跨层稀疏 KV 复用、四位全局缓存和近似窗口重放组合起来，把长输入的预填充主干计算近似减半、全局 KV 降至每 token 890 字节，并在发布方评测中显著提升多项 agent 能力；端到端服务收益和近似重放的极端条件可靠性仍缺少充分公开验证。
归档 更新
DSpark: Confidence Scheduled Speculative Decoding with Semi Autoregressive Generation
用 Markov head、置信度校准和硬件感知前缀调度，把并行 drafter 推进生产 serving。
待审阅 2026-06-27-dspark-confidence-scheduled-speculative-decoding Speculative DecodingMulti-Token PredictionServing Runtime
归档 更新
DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models
用细粒度专家切分和共享专家隔离提高 routed expert 的专门化程度。
归档 更新
Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models
用 hashed N gram lookup 和 context aware gating 增加可离线扩展的 conditional memory。
归档 更新
DeepSeek V3 Technical Report
把 auxiliary loss free MoE balancing、MTP、FP8 与 DualPipe 组合成高效大规模训练系统。
待审阅 2412.19437-deepseek-v3-technical-report MoE ArchitectureDistributed TrainingMulti-Token Prediction
归档 更新
Math Shepherd: Verify and Reinforce LLMs Step by step without Human Annotations
用 continuation 成功率自动生成 step level pseudo labels，训练 PRM 并接入 PPO。
待审阅 2312.08935-math-shepherd-automatic-process-supervision Process SupervisionVerifierReward Modeling