Skip to content

Supported Models

Cornserve supports a wide range of multimodal models from Hugging Face Hub, with flexible deployment options that enable efficient resource sharing and component disaggregation.

Overview

Cornserve's model support philosophy:

  • Hugging Face Native: Models are loaded directly from Hugging Face Hub using standard model IDs.
  • Flexible Deployment: Monolithic, encoder-fission, disaggregated, etc., you choose.
  • Component Sharing: Multiple models ("apps") can share encoder, decoder, or other component deployments if they are the same.
  • Multimodal First: Native support for text, image, video, and audio inputs and outputs.

Models by Category

Omni-modality (Any-to-Any) Models

Models that map multiple modalities to multiple modalities.

Model Family Example Models Input Modalities Output Modalities
Qwen3-Omni Qwen/Qwen3-Omni-30B-A3B-Instruct Text + Image + Video + Audio Text + Audio

Multimodal Input Models

Multimodal LLMs that accept multiple input modalities (e.g., text + image) and produce text outputs.

Model Family Example Models Input Modalities
Gemma 3 google/gemma-3-4b-it
google/gemma-3-12b-it
google/gemma-3-27b-it
Text + Image
Qwen2-VL Qwen/Qwen2-VL-7B-Instruct Text + Image + Video
Qwen2.5-VL Qwen/Qwen2.5-VL-7B-Instruct Text + Image + Video
Qwen3-VL Qwen/Qwen3-VL-4B-Instruct Text + Image + Video
Qwen3-VL-MoE Qwen/Qwen3-VL-30B-A3B-Instruct Text + Image + Video
InternVL3 OpenGVLab/InternVL3-1B
OpenGVLab/InternVL3-38B
Text + Image + Video
LLaVA-OneVision llava-hf/llava-onevision-qwen2-7b-ov-hf Text + Image + Video

Multimodal Output Models

Models that generate non-text outputs (e.g., images).

Model Family Example Models Input Modalities Output Modalities
Qwen-Image Qwen/Qwen-Image Text Image

Text-Only Language Models

Cornserve automatically supports any text-only LLMs supported by vLLM. Refer to vLLM's supported models documentation for the complete list.

Note

Cornserve uses a fork of vLLM, which can lag slightly behind the latest vLLM releases. If you need support for a newly released model, please file an issue on our repository, or we'd be happy to accept contributions.