M*: A Modular, Extensible, Serving System for Multimodal Models

• URL: http://arxiv.org/abs/2606.12688v2
• Date: Sat, 13 Jun 2026 05:57:49 GMT
• Title: M*: A Modular, Extensible, Serving System for Multimodal Models
• Abstract summary: We present M*, a universal serving system for efficient serving of composite AI models.<n>We show how it can concisely capture composite models from a broad range of families.<n>M* also outperforms the V-JEPA 2-AC rollout baseline for robotic planning by up to 12.5x.
• Score: 62.77975969000349
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We are entering a new era of composite model architectures that integrate diverse components such as vision encoders, language backbones, diffusion and flow heads, audio codecs, action generators, and world-model predictors. Such architectures underpin a broad class of multimodal models, including unified multimodal models, omni models, speech-language models, vision-language-action policies, and world models. However, existing model serving frameworks were built on narrow assumptions about model structure, making them ill-suited to accommodate this new architectural diversity. Here we present M*, a universal serving system for efficient serving of composite AI models. M* represents models as dataflow graphs, processing requests spanning diverse modalities and tasks as traversals over these graphs. The core insight is a modular abstraction that supports arbitrary composition of model components, flexible placement onto a physical cluster, and model-agnostic optimizations within a distributed runtime. We call this abstraction the Walk Graph and show how it can concisely capture composite models from a broad range of families. We instantiate M* on representative models and find that it achieves, on average, 20% lower end-to-end latency than vLLM-Omni for text-to-image workloads on BAGEL, while delivering up to 2.9x lower real-time factor and 2.7x higher throughput for text-to-speech workloads on Qwen3-Omni. M* also outperforms the V-JEPA 2-AC rollout baseline for robotic planning by up to 12.5x. Thus, our work paves the road towards more efficient serving of complex models with minimal developer effort.

Related papers

• Analyzing Diffusion and Autoregressive Vision Language Models in Multimodal Embedding Space [52.34072027212278]
  Embedding models are a fundamental component of modern AI systems such as semantic search and retrieval-augmented generation.<n>Recent advances in large foundation models have substantially accelerated the development of embedding models.<n>We present the first systematic study of converting Multimodal dLLMs into embedding models.
  arXiv  Detail & Related papers  (2026-01-19T06:51:15Z)
• Show-o2: Improved Native Unified Multimodal Models [57.34173415412808]
  Show-o2 is a native unified multimodal models that leverage autoregressive modeling and flow matching.<n>Built upon a 3D causal variational autoencoder space, unified visual representations are constructed through a dual-path of spatial (-temporal) fusion.
  arXiv  Detail & Related papers  (2025-06-18T15:39:15Z)
• OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging [124.91183814854126]
  Model merging seeks to combine multiple expert models into a single model.<n>We introduce a benchmark for model merging research that clearly divides the tasks for MLLM training and evaluation.<n>We find that model merging offers a promising way for building improved MLLMs without requiring training data.
  arXiv  Detail & Related papers  (2025-05-26T12:23:14Z)
• VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks [60.5257456681402]
  We study the potential for building universal embeddings capable of handling a wide range of downstream tasks.<n>We build a series of VLM2Vec models on SoTA VLMs like Phi-3.5-V, LLaVA-1.6 and evaluate them on MMEB's evaluation split.<n>Our results show that VLM2Vec achieves an absolute average improvement of 10% to 20% over existing multimodal embedding models.
  arXiv  Detail & Related papers  (2024-10-07T16:14:05Z)
• ACE: All-round Creator and Editor Following Instructions via Diffusion Transformer [40.32254040909614]
  We propose ACE, an All-round Creator and Editor, for visual generation tasks. We first introduce a unified condition format termed Long-context Condition Unit (LCU) We then propose a novel Transformer-based diffusion model that uses LCU as input, aiming for joint training across various generation and editing tasks.
  arXiv  Detail & Related papers  (2024-09-30T17:56:27Z)
• SEED-X: Multimodal Models with Unified Multi-granularity Comprehension and Generation [61.392147185793476]
  We present a unified and versatile foundation model, namely, SEED-X.<n>SEED-X is able to model multi-granularity visual semantics for comprehension and generation tasks.<n>We hope that our work will inspire future research into what can be achieved by versatile multimodal foundation models in real-world applications.
  arXiv  Detail & Related papers  (2024-04-22T17:56:09Z)
• A Lightweight Feature Fusion Architecture For Resource-Constrained Crowd Counting [3.5066463427087777]
  We introduce two lightweight models to enhance the versatility of crowd-counting models. These models maintain the same downstream architecture while incorporating two distinct backbones: MobileNet and MobileViT. We leverage Adjacent Feature Fusion to extract diverse scale features from a Pre-Trained Model (PTM) and subsequently combine these features seamlessly.
  arXiv  Detail & Related papers  (2024-01-11T15:13:31Z)
• Reformulating Vision-Language Foundation Models and Datasets Towards Universal Multimodal Assistants [65.47222691674074]
  Muffin framework employs pre-trained vision-language models to act as providers of visual signals. UniMM-Chat dataset explores the complementarities of datasets to generate 1.1M high-quality and diverse multimodal instructions.
  arXiv  Detail & Related papers  (2023-10-01T12:35:18Z)

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.