Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing

• URL: http://arxiv.org/abs/2512.17574v1
• Date: Fri, 19 Dec 2025 13:40:13 GMT
• Title: Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing
• Authors: Lingxiao Zhao, Haoran Zhou, Yuezhi Che, Dazhao Cheng,
• Abstract summary: Multimodal large language models (MLLMs) extend visual understanding through a three-stage pipeline.<n> multimodal preprocessing-especially video decoding-often dominates Time-to-First-Token (TTFT)<n>We present FlashCodec and UnifiedServe, two complementary designs that jointly optimize the end-to-end MLLM pipeline.
• Score: 16.063514680699576
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal large language models (MLLMs) extend LLMs with visual understanding through a three-stage pipeline: multimodal preprocessing, vision encoding, and LLM inference. While these stages enhance capability, they introduce significant system bottlenecks. First, multimodal preprocessing-especially video decoding-often dominates Time-to-First-Token (TTFT). Most systems rely on CPU-based decoding, which severely limits throughput, while existing GPU-based approaches prioritize throughput-oriented parallelism and fail to meet the latency-sensitive requirements of MLLM inference. Second, the vision encoder is a standalone, compute-intensive stage that produces visual embeddings and cannot be co-batched with LLM prefill or decoding. This heterogeneity forces inter-stage blocking and increases token-generation latency. Even when deployed on separate GPUs, these stages underutilize available compute and memory resources, reducing overall utilization and constraining system throughput. To address these challenges, we present FlashCodec and UnifiedServe, two complementary designs that jointly optimize the end-to-end MLLM pipeline. FlashCodec accelerates the multimodal preprocessing stage through collaborative multi-GPU video decoding, reducing decoding latency while preserving high throughput. UnifiedServe optimizes the vision-to-text and inference stages using a logically decoupled their execution to eliminate inter-stage blocking, yet physically sharing GPU resources to maximize GPU system utilization. By carefully orchestrating execution across stages and minimizing interference, UnifiedServe Together, our proposed framework forms an end-to-end optimized stack that can serve up to 3.0$\times$ more requests or enforce 1.5$\times$ tighter SLOs, while achieving up to 4.4$\times$ higher throughput compared to state-of-the-art systems.

Related papers

• Spava: Accelerating Long-Video Understanding via Sequence-Parallelism-aware Approximate Attention [63.69228529380251]
  Spava is a sequence-parallel framework with optimized attention for long-video inference.<n>Spava delivers speedups of 12.72x, 1.70x, and 1.18x over FlashAttn, ZigZagRing, and APB, without notable performance loss.
  arXiv  Detail & Related papers  (2026-01-29T09:23:13Z)
• VidLaDA: Bidirectional Diffusion Large Language Models for Efficient Video Understanding [52.69880888587866]
  Current Video Large Language Models (Video LLMs) typically encode frames via a encoder vision and employ an autoregressive (AR) LLM for understanding and generation.<n>We propose VidLaDA, a Diffusion Video LLM based on Language Models (DLMs) that leverages bidirectional attention to unlock comprehensive modeling and decode tokens in parallel.<n>Experiments show VidLaDA rivals state-of-the-art AR baselines and outperforms DLM baselines, with MARS-Cache delivering over 12x speedup without compromising accuracy.
  arXiv  Detail & Related papers  (2026-01-25T15:02:01Z)
• DuetServe: Harmonizing Prefill and Decode for LLM Serving via Adaptive GPU Multiplexing [15.376910065679994]
  DuetServe is a unified LLM serving framework that achieves disaggregation-level isolation within a single GPU.<n>DuetServe improves total throughput by up to 1.3x while maintaining low generation latency compared to state-of-the-art frameworks.
  arXiv  Detail & Related papers  (2025-11-06T20:18:34Z)
• Eliminating Multi-GPU Performance Taxes: A Systems Approach to Efficient Distributed LLMs [61.953548065938385]
  We introduce the ''Three Taxes'' (Bulk Synchronous, Inter- Kernel Data Locality, and Kernel Launch Overhead) as an analytical framework.<n>We propose moving beyond the rigid BSP model to address key inefficiencies in distributed GPU execution.<n>We observe a 10-20% speedup in end-to-end latency over BSP-based approaches.
  arXiv  Detail & Related papers  (2025-11-04T01:15:44Z)
• Free-MoRef: Instantly Multiplexing Context Perception Capabilities of Video-MLLMs within Single Inference [88.57742986765238]
  Free-MoRef is a training-free approach to multiplex the context perception capabilities of Video-MLLMs.<n>Experiments show that Free-MoRef achieves full perception of 2$times$ to 8$times$ longer input frames without compression on a single A100 GPU.
  arXiv  Detail & Related papers  (2025-08-04T07:31:10Z)
• Nexus:Proactive Intra-GPU Disaggregation of Prefill and Decode in LLM Serving [4.309392302169281]
  Engine-level prefill-decode (PD) disaggregation avoids interference but incurs higher hardware and coordination overhead.<n>PD achieves up to 2.2x higher throughput, 20x lower TTFT, and 2.5x lower TBT than vLLM; outperforms SG by up to 2x; and matches or exceeds disaggregated vLLM.
  arXiv  Detail & Related papers  (2025-07-09T07:27:18Z)
• 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)
• FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving [9.386969461835433]
  FlashInfer is a customizable and efficient attention engine for large language models (LLMs)<n>It tackles KV-cache storage heterogeneity using block-sparse format and composable formats to optimize memory access and reduce redundancy.<n>It also offers a customizable attention template, enabling adaptation to various settings through Just-In-TimeJIT compilation.
  arXiv  Detail & Related papers  (2025-01-02T02:02:20Z)
• FusionAI: Decentralized Training and Deploying LLMs with Massive Consumer-Level GPUs [57.12856172329322]
  We envision a decentralized system unlocking the potential vast untapped consumer-level GPU. This system faces critical challenges, including limited CPU and GPU memory, low network bandwidth, the variability of peer and device heterogeneity.
  arXiv  Detail & Related papers  (2023-09-03T13:27:56Z)

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.