arXiv cs.CL· Mingkang Liu, Weiming Meng, Peipei Jiao, Jiawei Zhang, Tianyuan Liu, Junyan Li, Lichen Pang·· 13 小时前AI 评分26
RCSD：面向领域特定投机解码的强化与持续学习框架
RCSD: Reinforcement and Continual Learning for Domain-Specific Speculative Decoding
AI 导读
RCSD 提出一种投机解码训练框架，用推理中自然产生的接受与拒绝序列构建隐式偏好对，并配合掩码损失避免惩罚与目标模型对齐的 token，从而直接优化接受长度。在多个基准上，其平均接受长度比 HASS 提升 35%，实际负载下端到端推理速度达 HASS 的 1.35 倍，且不损害生成质量。
正文 · 原文
Abstract:Speculative decoding accelerates large language model inference by employing a compact draft model to propose candidate tokens that are verified in parallel by a target model. Conventional training relies on soft-label distillation, which forces the draft model to mimic the target's full output distribution, including high-entropy regions that degrade acceptance rates and increase early rejection risk. We propose RCSD (Reinforcement and Continual Learning for Speculative Decoding), a training framework that aligns draft model optimization with the system-level objective of maximizing acceptance length. RCSD leverages naturally occurring accepted and rejected sequences during speculative inference to construct implicit preference pairs, with masked loss to avoid penalizing target-aligned tokens. Evaluated across multiple benchmarks, RCSD improves average acceptance length by 35% over HASS and achieves 1.35x faster end-to-end inference than HASS on practical workloads, without compromising generation quality.
| Comments: | 9 pages, 3 figures |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.29527 [cs.CL] |
| (or arXiv:2610.29527v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29527 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Lichen Pang [view email]
[v1]
Wed, 21 Oct 2026 10:16:23 UTC (685 KB)
来源：arXiv cs.CL · arxiv.org