When RL Meets Adaptive Speculative Training: A Unified Training-Serving System
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866914568426487808 |
|---|---|
| author | Wang, Junxiong Bie, Fengxiang Li, Jisen Zhou, Zhongzhu Shao, Zelei Wang, Yubo Liu, Yinghui Wu, Qingyang May, Avner Yanamandra, Sri Zhang, Ce Dao, Tri Liang, Percy Athiwaratkun, Ben Song, Shuaiwen Leon Xu, Chenfeng Wu, Xiaoxia |
| author_facet | Wang, Junxiong Bie, Fengxiang Li, Jisen Zhou, Zhongzhu Shao, Zelei Wang, Yubo Liu, Yinghui Wu, Qingyang May, Avner Yanamandra, Sri Zhang, Ce Dao, Tri Liang, Percy Athiwaratkun, Ben Song, Shuaiwen Leon Xu, Chenfeng Wu, Xiaoxia |
| contents | Speculative decoding can significantly accelerate LLM serving, yet most deployments today disentangle speculator training from serving, treating speculator training as a standalone offline modeling problem. We show that this decoupled formulation introduces substantial deployment and adaptation lag: (1) high time-to-serve, since a speculator must be trained offline for a considerable period before deployment; (2) delayed utility feedback, since the true end-to-end decoding speedup is only known after training and cannot be inferred reliably from acceptance rate alone due to model-architecture and system-level overheads; and (3) domain-drift degradation, as the target model is repurposed to new domains and the speculator becomes stale and less effective.
To address these issues, we present Aurora, a unified training-serving system that closes the loop by continuously learning a speculator directly from live inference traces. Aurora reframes online speculator learning as an asynchronous reinforcement-learning problem: accepted tokens provide positive feedback, while rejected speculator proposals provide implicit negative feedback that we exploit to improve sample efficiency. Our design integrates an SGLang-based inference server with an asynchronous training server, enabling hot-swapped speculator updates without service interruption. Crucially, Aurora supports day-0 deployment: a speculator can be served immediately and rapidly adapted to live traffic, improving system performance while providing immediate utility feedback. Across experiments, Aurora achieves a 1.5x day-0 speedup on recently released frontier models (e.g., MiniMax M2.1 229B and Qwen3-Coder-Next 80B). Aurora also adapts effectively to distribution shifts in user traffic, delivering an additional 1.25x speedup over a well-trained but static speculator on widely used models (e.g., Qwen3 and Llama3). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_06932 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | When RL Meets Adaptive Speculative Training: A Unified Training-Serving System Wang, Junxiong Bie, Fengxiang Li, Jisen Zhou, Zhongzhu Shao, Zelei Wang, Yubo Liu, Yinghui Wu, Qingyang May, Avner Yanamandra, Sri Zhang, Ce Dao, Tri Liang, Percy Athiwaratkun, Ben Song, Shuaiwen Leon Xu, Chenfeng Wu, Xiaoxia Machine Learning Speculative decoding can significantly accelerate LLM serving, yet most deployments today disentangle speculator training from serving, treating speculator training as a standalone offline modeling problem. We show that this decoupled formulation introduces substantial deployment and adaptation lag: (1) high time-to-serve, since a speculator must be trained offline for a considerable period before deployment; (2) delayed utility feedback, since the true end-to-end decoding speedup is only known after training and cannot be inferred reliably from acceptance rate alone due to model-architecture and system-level overheads; and (3) domain-drift degradation, as the target model is repurposed to new domains and the speculator becomes stale and less effective. To address these issues, we present Aurora, a unified training-serving system that closes the loop by continuously learning a speculator directly from live inference traces. Aurora reframes online speculator learning as an asynchronous reinforcement-learning problem: accepted tokens provide positive feedback, while rejected speculator proposals provide implicit negative feedback that we exploit to improve sample efficiency. Our design integrates an SGLang-based inference server with an asynchronous training server, enabling hot-swapped speculator updates without service interruption. Crucially, Aurora supports day-0 deployment: a speculator can be served immediately and rapidly adapted to live traffic, improving system performance while providing immediate utility feedback. Across experiments, Aurora achieves a 1.5x day-0 speedup on recently released frontier models (e.g., MiniMax M2.1 229B and Qwen3-Coder-Next 80B). Aurora also adapts effectively to distribution shifts in user traffic, delivering an additional 1.25x speedup over a well-trained but static speculator on widely used models (e.g., Qwen3 and Llama3). |
| title | When RL Meets Adaptive Speculative Training: A Unified Training-Serving System |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2602.06932 |