paper-with-me

Papers

RW-TTT: Batched Serving for Request-Owned Test-Time Training State

2026-05-27 · Jian Yang, Zhizhuo Kou, Yao Tian, Hao Zhang, Han Chen, Sirui Han, Yike Guo arxiv

Test-time training (TTT) adapts an LLM during generation by reading and updating request-owned state, such as fast weights, low-rank deltas, or streaming learner state. This breaks batched LLM serving, which assumes shared static weights: serial execution is correct but slow, while naive batching can corrupt request state. We formulate this problem as read-write TTT serving and present RW-TTT , which tags each decode step with its owner, version, and READ/WRITE effect, batches only compatible phases, and commits updates only to the owner. On one GPU with eight fast-weight InPlace-TTT streams, RW-TTT reaches 274.61 aggregate tok/s, 9.31x over sequential serving and 3.44x over per-stream replicas under the same memory budget. It preserves behavior on RULER, a long-context benchmark, and passes owner/version checks.

📄 PDF Abstract BibTeX arXiv:2605.28053

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

PackInfer: Compute- and I/O-Efficient Attention for Batched LLM Inference

2026-02-03 · Rui Ning, Wei Zhang, Fan Lai arxiv

Attention efficiency is critical to large language model (LLM) inference. While prior advances optimize attention execution for individual requests (e.g., FlashAttention), production LLM serving relies on batching reques…

SharedRequest: Privacy-Preserving Model-Agnostic Inference for Large Language Models

2026-06-03 · Peihua Mai, Xuanrong Gao, Youlong Ding, Xianglong Du 외 arxiv

With the widespread deployment of public large language models (LLMs) such as ChatGPT, protecting user prompt privacy has become an increasingly critical issue. Existing privacy-preserving inference methods sacrifice eit…

Regulating Branch Parallelism in LLM Serving

2026-05-07 · Swapnil Gandhi, Siva Hari, William J. Dally, Christos Kozyrakis arxiv

Recent methods expose intra-request parallelism in LLM outputs, allowing independent branches to decode concurrently. Existing serving systems execute these branches eagerly or under fixed caps. We show that both are bri…

CARVE: Certified Affordable Repair of Vetoed Maneuvers via Envelopes for Interactive Driving

2026-05-31 · Yifan Wang arxiv

Interactive driving exposes a failure mode that is easy to miss in rule-aware autonomous-driving stacks: a hard-rule margin can be negative for an ego candidate even though a small lawful accommodation by a non-priority …

LMSM: LLM Security Framework Inspired by Linux Security Modules

2026-08-26 · XiuYu Zhang, Bonan Ruan, Junfeng Fang, An Zhang 외 hf

Large language models (LLMs) are increasingly deployed with layered defenses, yet malicious prompts can still bypass them. Interpretability methods can expose model-internal signals along the generation path that could i…