The most rapid route to a local installation of this model is through Docker.
Use the instructions provided below to complete the setup.
The installer auto-downloads and deploys the entire model pack.
Once launched, the setup wizard will detect your specs to configure the model for maximum efficiency.
Qwen3-VL-Embedding-2B is a compact yet powerful multimodal embedding model that processes text, images, and videos into a unified vector space. It leverages a vision-language transformer architecture with 2 billion parameters, delivering state‑of‑the‑art retrieval performance across diverse benchmarks. The model supports high‑resolution visual inputs and can handle up to 2048‑token text sequences, enabling flexible downstream tasks such as image search and cross‑modal retrieval. Its training pipeline incorporates large‑scale paired datasets, ensuring robust semantic alignment between modalities while maintaining computational efficiency. The resulting embeddings are widely adopted in production systems due to their fast inference and low memory footprint.
| Spec | Value |
|---|---|
| Parameters | 2 B |
| Embedding Dim | 1024 |
| Supported Modalities | Text, Image, Video |
| Max Text Tokens | 2048 |
| Max Image Resolution | 1024×1024 |
