Rocks, Pebbles and Sand: Modality-aware Scheduling for Multimodal Large Language Model Inference

• URL: http://arxiv.org/abs/2603.26498v1
• Date: Fri, 27 Mar 2026 15:00:53 GMT
• Title: Rocks, Pebbles and Sand: Modality-aware Scheduling for Multimodal Large Language Model Inference
• Abstract summary: Multimodal Large Language Models (MLLMs) power platforms like ChatGPT, Gemini, and Copilot, enabling richer interactions with text, images, and videos.<n>Existing LLM serving systems fail under multimodality: large requests monopolize resources, causing severe head-of-line blocking and performance degradation.<n>We design RPS-Serve, a modality-aware scheduler that lets sand flow quickly through pebbles and rocks, ensuring interactive responsiveness while avoiding starvation.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal Large Language Models (MLLMs) power platforms like ChatGPT, Gemini, and Copilot, enabling richer interactions with text, images, and videos. These heterogeneous workloads introduce additional inference stages, such as vision preprocessing and encoding, that inflate latency and memory demand. Existing LLM serving systems, optimized for text-only workloads, fail under multimodality: large requests (e.g., videos) monopolize resources, causing severe head-of-line blocking and performance degradation. Our key insight is that multimodal requests differ by orders of magnitude in resource demands, which we capture through a simple abstraction: videos behave like rocks, images like pebbles, and text like sand. We design RPS-Serve, a modality-aware scheduler that lets sand flow quickly through pebbles and rocks, ensuring interactive responsiveness while avoiding starvation. RPS-Serve classifies requests, prioritizes them dynamically, and applies aging to avoid starvation. Evaluation across state-of-the-art MLLMs shows that RPS-Serve reduces, on average, time-to-first-token (TTFT) by 54% overall, and by 78.5% for latency-critical requests, compared to current systems. RPS-Serve delivers LLM-like responsiveness for MLLMs, with modality-aware scheduling and by making the most efficient use of the available resources.

Related papers

• StreamWise: Serving Multi-Modal Generation in Real-Time at Scale [7.73695790907204]
  Multi-modal generative models are enabling new applications, from storytelling to automated media synthesis.<n>Currently, serving real-time multi-modal at scale is costly and complex, requiring efficient coordination of diverse models.<n>We design an adaptive, modular serving system, StreamWise, that dynamically manages quality (e.g., resolution, sharpness), model/content parallelism, and resource-aware scheduling.
  arXiv  Detail & Related papers  (2026-03-06T01:22:16Z)
• Refer-Agent: A Collaborative Multi-Agent System with Reasoning and Reflection for Referring Video Object Segmentation [50.22481337087162]
  Referring Video Object (RVOS) aims to segment objects in videos based on textual queries.<n>Refer-Agent is a collaborative multi-agent system with alternating reasoning-reflection mechanisms.
  arXiv  Detail & Related papers  (2026-02-03T14:48:12Z)
• A Comprehensive Study on Visual Token Redundancy for Discrete Diffusion-based Multimodal Large Language Models [85.30893355216486]
  We study how visual token redundancy evolves with different dMLLM architectures and tasks.<n>Our study reveals that visual redundancy emerges only in from-scratch dMLLMs while handling long-answer tasks.<n>Layer-skipping is promising for accelerating AR-to-diffusion dMLLMs, whereas progressive or late-step pruning is more effective for from-scratch dMLLMs.
  arXiv  Detail & Related papers  (2025-11-19T04:13:36Z)
• ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models [67.75439511654078]
  Large Vision-Language Models (LVLMs) have introduced a new paradigm for understanding and reasoning about image input through textual responses.<n>They face the persistent challenge of hallucination, which introduces practical weaknesses and raises concerns about their reliable deployment in real-world applications.<n>We propose ONLY, a training-free decoding approach that requires only a single query and a one-layer intervention during decoding, enabling efficient real-time deployment.
  arXiv  Detail & Related papers  (2025-07-01T16:01:08Z)
• Tempo: Application-aware LLM Serving with Mixed SLO Requirements [7.290735867969561]
  We introduce Tempo, a scheduler designed to maximize service gain across diverse LLM workloads.<n>Our evaluation shows that Tempo improves end-to-end service gain by up to 8.3$times$ achieves and up to 10.3$times$ SLO goodput compared to state-of-the-art designs.
  arXiv  Detail & Related papers  (2025-04-24T05:55:21Z)
• STORM: Token-Efficient Long Video Understanding for Multimodal LLMs [116.4479155699528]
  STORM is a novel architecture incorporating a dedicated temporal encoder between the image encoder and the Video-LLMs.<n>We show that STORM achieves state-of-the-art results across various long video understanding benchmarks.
  arXiv  Detail & Related papers  (2025-03-06T06:17:38Z)
• DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution [114.61347672265076]
  Development of MLLMs for real-world robots is challenging due to the typically limited computation and memory capacities available on robotic platforms. We propose a Dynamic Early-Exit Framework for Robotic Vision-Language-Action Model (DeeR) that automatically adjusts the size of the activated MLLM. DeeR demonstrates significant reductions in computational costs of LLM by 5.2-6.5x and GPU memory of LLM by 2-6x without compromising performance.
  arXiv  Detail & Related papers  (2024-11-04T18:26:08Z)
• Queue management for slo-oriented large language model serving [3.0134961904579094]
  We propose QLM, a queue management system for large language model (LLM) serving.<n>QLM maintains batch and interactive requests across different models and SLOs in a request queue.<n>It uses a Request Waiting Time (RWT) Estimator that estimates the waiting times for requests in the request queue.
  arXiv  Detail & Related papers  (2024-06-05T21:17:34Z)
• Llumnix: Dynamic Scheduling for Large Language Model Serving [17.919408899409113]
  Inference serving for large language models (LLMs) is the key to unleashing their potential. We introduce Llumnix, an LLM serving system that reacts to such heterogeneous and unpredictable requests by runtime rescheduling. We show that Llumnix improves tail latencies by an order of magnitude, accelerates high-priority requests by up to 1.5x, and delivers up to 36% cost savings.
  arXiv  Detail & Related papers  (2024-06-05T13:20:18Z)
• Efficient Interactive LLM Serving with Proxy Model-based Sequence Length Prediction [8.705908108054878]
  Large models (LLMs) have been driving a new wave of AI applications across numerous domains. We present a speculative shortest-job-first (SSJF) scheduler that uses a light proxy model to predict LLM output sequence lengths.
  arXiv  Detail & Related papers  (2024-04-12T14:46:15Z)
• CALF: Aligning LLMs for Time Series Forecasting via Cross-modal Fine-Tuning [59.88924847995279]
  We propose a novel Cross-Modal LLM Fine-Tuning (CALF) framework for MTSF.<n>To reduce the distribution discrepancy, we develop the cross-modal match module.<n>CALF establishes state-of-the-art performance for both long-term and short-term forecasting tasks.
  arXiv  Detail & Related papers  (2024-03-12T04:04:38Z)
• Fast Distributed Inference Serving for Large Language Models [12.703624317418237]
  We present FastServe, a distributed inference serving system for large language models (LLMs) FastServe exploits the autoregressive pattern of LLM inference to enable preemption at the granularity of each output token. We build a system prototype of FastServe and experimental results show that compared to the state-of-the-art solution vLLM, FastServe improves the throughput by up to 31.4x and 17.9x under the same average and tail latency requirements, respectively.
  arXiv  Detail & Related papers  (2023-05-10T06:17:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.