14 papers
Router Drift Control: Mitigating Off-Policy Degradation in Mixture-of-Experts Reinforcement Learning
Liangyao Jin, Zhixin Wang, Liming Liu +8
Reinforcement learning (RL) can become off-policy due to training-inference mismatch, mini-batch updates, and asynchronous training. The resulting mismatch is especially harmful in…
Coda: Exploiting Admission Flexibility for Coding-Agent Serving
Youhe Jiang, Fangcheng Fu, Binhang Yuan +5
Coding agents powered by large language models (LLMs) repeatedly alternate between model inference and tool calls, creating long-lived sessions with reusable key-value (KV) states…
Denoising Surface: Modeling and Predicting Inference Cost for Diffusion LLM Serving
Haoyu Zheng, Fangcheng Fu, Binhang Yuan +6
As diffusion large language models (dLLMs) become more capable, they are moving from research settings to real-world \textit{serving}, where request management (such as scheduling…
Efficient Agentic LLM Serving over SSD-based Sparse KV Storage
Wenhao He, Ping Zhang, Xiaohe Hu +5
Agentic sessions driven by Large language models (LLMs) often alternate between model inference and tool use, accumulating long histories across successive rounds. Serving these se…
Vidu S1: A Real-Time Interactive Video Generation Model
Jintao Zhang, Kai Jiang, Jintao Chen +24
We introduce Vidu S1, a real-time interactive video generation model supporting voice control of digital characters. Users can control video generation content at any moment throug…
Demystifying the Design Space and Best Practices for Heterogeneous LLM Inference and Serving
Zhixin Wang, Zhengbo Wang, Fangcheng Fu +15
Heterogeneous prefill-decode (PD) inference is now in production: prefill on cost-efficient or supply-available accelerators, decode on bandwidth-strong ones, and KV state crossing…
TurboServe: Serving Streaming Video Generation Efficiently and Economically
Youhe Jiang, Haoxu Wang, Haotong Bao +5
Streaming video generation is emerging as a new serving workload in which users interact with long-lived sessions that generate video progressively, chunk by chunk. Unlike offline…
Efficient Serving for Dynamic Agent Workflows with Prediction-based KV-Cache Management
Haoyu Zheng, Fangcheng Fu, Jia Wu +6
LLM-based workflows compose specialized agents to execute complex tasks, and these agents usually share substantial context, allowing KV-Cache reuse to save computation. Existing a…
Autopoiesis: A Self-Evolving System Paradigm for LLM Serving Under Runtime Dynamics
Youhe Jiang, Ran Yan, You Peng +4
Modern Large Language Model (LLM) serving operates in highly volatile environments characterized by severe runtime dynamics, such as workload fluctuations and elastic cluster autos…
BOute: Cost-Efficient LLM Serving with Heterogeneous LLMs and GPUs via Multi-Objective Bayesian Optimization
Youhe Jiang, Fangcheng Fu, Eiko Yoneki
The rapid growth of large language model (LLM) deployments has made cost-efficient serving systems essential. Recent efforts to enhance system cost-efficiency adopt two main perspe…
Efficient Multi-round LLM Inference over Disaggregated Serving
Wenhao He, Youhe Jiang, Penghao Zhao +4
With the rapid evolution of Large Language Models (LLMs), multi-round workflows, such as autonomous agents and iterative retrieval, have become increasingly prevalent. However, thi…
OServe: Accelerating LLM Serving via Spatial-Temporal Workload Orchestration
Youhe Jiang, Fangcheng Fu, Taiyi Wang +2
Serving Large Language Models (LLMs) can benefit immensely from parallelizing both the model and input requests across multiple devices, but incoming workloads exhibit substantial…
LMDeploy Accelerates Mixed-Precision LLM Inference with TurboMind
Li Zhang, Youhe Jiang, Guoliang He +6
Mixed-precision inference techniques reduce the memory and computational demands of Large Language Models (LLMs) by applying hybrid precision formats to model weights, activations,…
Cascadia: An Efficient Cascade Serving System for Large Language Models
Youhe Jiang, Fangcheng Fu, Wanru Zhao +4
Recent advances in large language models (LLMs) have intensified the need to deliver both rapid responses and high-quality outputs. More powerful models yield better results but in…