Yu Wu (吴俣) | Chlience Paper Archive

归档 Jun 28, 2026

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 Decoding Multi-Token Prediction Serving Runtime

归档 Jun 24, 2026

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。

待审阅 2401.06066-deepseekmoe-expert-specialization MoE Architecture

归档 Jun 23, 2026

把 auxiliary loss free MoE balancing、MTP、FP8 与 DualPipe 组合成高效大规模训练系统。

归档 Jun 23, 2026

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 Supervision Verifier Reward Modeling

归档 Feb 06, 2026

用 outcome based RL 从强 base model 诱导 long CoT、自验证与策略切换，再通过 SFT / RL / distillation 转为可用模型。

待审阅 2501.12948-deepseek-r1-rl-reasoning Reasoning RL Reward Modeling RL Algorithm

Related Notes