Yuxuan Tong (童雨轩)

TRM and OTB coauthor, recurring verl/DAPO infrastructure contributor, and Laminar equal-contribution author. Homepage and GitHub identify him as Shawn/Yuxuan Tong, Tsinghua DCST student and ByteDance Seed AI Researcher; GitHub links the same homepage and X account.

4 archived notes X: high Homepage GitHub X

归档 Jun 28, 2026

Laminar: A Scalable Asynchronous RL Post Training Framework

让完成轨迹独立进入 experience buffer，并以 CPU/RDMA relay 和同版本 repack 解除全局 batch / 权重同步 barrier。

待审阅 2510.12633-laminar-asynchronous-rl-post-training RL Infrastructure Rollout Optimization Distributed Training

归档 Jun 22, 2026

The Optimal Token Baseline: Variance Reduction for Long Horizon LLM RL

推导 token level variance minimizing baseline，并用 logit gradient proxy 近似长轨迹 policy gradient 权重。

待审阅 2602.07078-optimal-token-baseline-long-horizon-llm-rl Credit Assignment RL Algorithm RL Theory

归档 Jun 21, 2026

Trust Region Masking for Long Horizon LLM Reinforcement Learning

用序列级 max token divergence 约束 rollout policy mismatch，控制长轨迹 surrogate error。

待审阅 2512.23075-trust-region-masking-long-horizon-llm-rl Training Stability RL Algorithm Rollout Optimization

归档 Mar 03, 2026

DAPO: An Open Source LLM Reinforcement Learning System at Scale

用 Clip Higher、Dynamic Sampling、token level loss 与 overlong shaping 构成可复现 long CoT RL recipe。

待审阅 2503.14476-dapo-long-cot-rl-system RL Algorithm Reasoning RL RL Infrastructure

Related Notes

Laminar: A Scalable Asynchronous RL Post Training Framework

The Optimal Token Baseline: Variance Reduction for Long Horizon LLM RL

Trust Region Masking for Long Horizon LLM Reinforcement Learning

DAPO: An Open Source LLM Reinforcement Learning System at Scale