tracked author

Junyang Lin

Qwen technical lead and recurring Qwen2.5/Qwen3/Bebop author. OpenReview records Principal Researcher at Qwen Team, Alibaba Group; homepage identifies Junyang Lin and recent work on language models, multimodal systems and agentic learning; X handle matches the homepage alias and posts on model reasoning/agentic thinking.

3 archived notes X: high Homepage X

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 Jul 09, 2026

Qwen3 Coder Next Technical Report

Qwen3 Coder Next 在 80B 总参数中每步激活 3B 参数，并用可执行仓库任务、长上下文中训、多模板工具训练、软件工程强化学习与专家蒸馏构建编码智能体，在三种 SWE Bench Verified scaffold 上达到 70.6% 至 71.3% 解决率，报告尚未拆分各训练组件的净贡献。

待审阅 2603.00729-qwen3-coder-next-agentic-coding Coding Agent Agent RL Tool Use

归档 Jun 23, 2026

Qwen2.5 Technical Report

把 18T 预训练、SFT / DPO / GRPO 与长上下文扩展组织为通用、代码和数学模型族。

待审阅 2412.15115-qwen2-5-technical-report Reasoning RL Long Context Coding Agent

归档 Jun 10, 2026

Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Sampling

以 TV loss 训练 MTP heads，并用 probabilistic rejection sampling 提高 RL rollout 接受率。

待审阅 2606.12370-bebop-mtp-rejection-sampling-rl-training Rollout Optimization Speculative Decoding Multi-Token Prediction