14 papers
Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist
Yuanhao Ban, Tong Xie, Sohyun An +6
Faithfulness -- how precisely a generated image aligns with its prompt -- is increasingly central to the real-world utility of text-to-image (T2I) models. Existing faithfulness ben…
BlendServe: Optimizing Offline Inference for Auto-regressive Large Models with Resource-aware Batching
Yilong Zhao, Shuo Yang, Kan Zhu +5
Offline batch inference, which leverages the flexibility of request batching to achieve higher throughput and lower costs, is becoming more popular for latency-insensitive applicat…
Continuum: Efficient and Robust Multi-Turn LLM Agent Scheduling with KV Cache Time-to-Live
Hanchen Li, Runyuan He, Qiuyang Mang +7
KV cache management is essential for efficient LLM inference. To maximize utilization, existing inference engines evict finished requests' KV cache if new requests are waiting. Thi…
RedunCut: Measurement-Driven Sampling and Accuracy Performance Modeling for Low-Cost Live Video Analytics
Gur-Eyal Sela, Kumar Krishna Agrawal, Bharathan Balaji +2
Live video analytics (LVA) runs continuously across massive camera fleets, but inference cost with modern vision models remains high. To address this, dynamic model size selection…
Towards Efficient and Practical GPU Multitasking in the Era of LLM
Jiarong Xing, Yifan Qiao, Simon Mo +5
GPU singletasking is becoming increasingly inefficient and unsustainable as hardware capabilities grow and workloads diversify. We are now at an inflection point where GPUs must em…
TurboSpec: Closed-loop Speculation Control System for Optimizing LLM Serving Goodput
Xiaoxuan Liu, Jongseok Park, Langxiang Hu +10
Large Language Model (LLM) serving systems batch concurrent user requests to achieve efficient serving. However, in real-world deployments, such inter-request parallelism from batc…