章程瑞东
Employers on the program
- 2026阿里巴巴1 talk
Events1
Conferences1
Talks1
Years active2026
The year around now
Talks in the past 12 months and announced talks ahead, one row per conference, January to December. Hover a mark for the talk.
Appearances by year
- AICon 全球人工智能开发与应用大会 (AICon Global AI Development and Application Conference)
Talks
| Date | Event | Conference | Role | Title | Links |
|---|---|---|---|---|---|
| Aug 21, 2026 | AICon 全球人工智能开发与应用大会 | 深圳 | AICon 全球人工智能开发与应用大会 (AICon Global AI Development and Application Conference) | speaker | 面向 Qwen 系列模型线性注意力的高性能优化实践 as 阿里巴巴 自 Qwen-3.5 采用混合注意力结构以来，Qwen 系列模型的线性注意力层 Gated Delta Network（GDN）在训练与推理中的开销日益显著。现有实现面临两大难题：一是多步算子需要反复读写 HBM，访存瓶颈严重；二是状态递推的串行性使算子并行度受限于注意力头数，在长序列、小 batch 或 TP 场景下…Show less自 Qwen-3.5 采用混合注意力结构以来，Qwen 系列模型的线性注意力层 Gated Delta Network（GDN）在训练与推理中的开销日益显著。现有实现面临两大难题：一是多步算子需要反复读写 HBM，访存瓶颈严重；二是状态递推的串行性使算子并行度受限于注意力头数，在长序列、小 batch 或 TP 场景下 GPU 利用率低下。为此，我们提出 FlashQLA —— 基于 TileLang 的高性能线性注意力算子库。核心方案包括：将 GDN Chunked Prefill 的前向与反向流程各拆分为两个融合算子，通过数据复用降低访存开销，并在算… |
Frequent co-speakers
No one else spoke at two or more of the same events.
Speakers with the same name are merged into one entry, so this page may combine more than one person. Affiliations are shown as each conference listed them.