arXiv cs.AI· Eljas Linna, Hao Dong, Juho Kanniainen·· 1 天前AI 评分34
RESYNC：面向截止时间约束流式推理的残差隐状态同步
RESYNC: Residual Latent-State Synchronization for Deadline-Bound Streaming Inference
AI 导读
RESYNC 让大模型异步运行，用快速路径以新观测更新其陈旧隐状态并复用原预测头，在 12 组数据集-架构组合上均优于最佳紧凑或蒸馏模型。在 1 ms 到达的实时回放中，同步大模型无法按时输出任何预测，RESYNC 单 CPU 线程下至少 99.4% 的观测在间隔内完成，中位延迟 0.35–0.62 ms，比同步大模型推理快 7.9–17.7 倍。
正文 · 原文
Abstract:Predicting every observation in a high-rate stream requires a model to finish before the next arrival. A large model may offer better predictions but miss that deadline. We propose RESYNC: the large model runs asynchronously, while a fast path updates its latest exact but stale latent state using the observations received since that state was computed and applies its unchanged prediction head. The updater learns from the frozen model's representations without task labels. Across one real and two synthetic streams and five architecture families, RESYNC beats the best compact or distilled alternative with a paired 95% interval above zero in all twelve dataset--architecture pairs with a material teacher advantage at anchor ages of 4--12 observations, and in eleven pairs at older tested ages up to 32. The share of the teacher's gain preserved remains complete on the real stream but is lower and generally declines with anchor age on the synthetic streams. In live replay with 1 ms arrivals, the synchronous large models cannot deliver any predictions on time while RESYNC serves at least 99.4% of arrivals within the interval, with 0.35--0.62 ms median latency on one CPU thread: around the same speed as or faster than the compact model, and 7.9--17.7$\times$ faster than synchronous large-model inference. RESYNC targets settings with background compute available but a strict per-observation response deadline.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.16523 [cs.LG] |
| (or arXiv:2611.16523v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.16523 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Eljas Linna [view email]
[v1]
Thu, 12 Nov 2026 09:19:39 UTC (431 KB)
来源：arXiv cs.AI · arxiv.org