jina-vlm: Small Multilingual Vision Language Model
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915977602531328 |
|---|---|
| author | Koukounas, Andreas Mastrapas, Georgios Hönicke, Florian Eslami, Sedigheh Roncari, Guillaume Martens, Scott Xiao, Han |
| author_facet | Koukounas, Andreas Mastrapas, Georgios Hönicke, Florian Eslami, Sedigheh Roncari, Guillaume Martens, Scott Xiao, Han |
| contents | We present jina-vlm, a token-efficient 2.4B parameter vision-language model that achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs. The model couples a SigLIP2 vision encoder with a Qwen3 language decoder and makes use of image tiling and attention-pooling for token-efficient processing of arbitrary-resolution images. To understand the contribution of different training data categories, we conduct a leave-one-out data mixture ablation study-systematically removing task, domain, modality, and language categories-to diagnose which data types are necessary versus redundant and whether task benefits transfer across domains. Model weights and code are publicly released at https://huggingface.co/jinaai/jina-vlm. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_04032 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | jina-vlm: Small Multilingual Vision Language Model Koukounas, Andreas Mastrapas, Georgios Hönicke, Florian Eslami, Sedigheh Roncari, Guillaume Martens, Scott Xiao, Han Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition 68T50 I.2.7 We present jina-vlm, a token-efficient 2.4B parameter vision-language model that achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs. The model couples a SigLIP2 vision encoder with a Qwen3 language decoder and makes use of image tiling and attention-pooling for token-efficient processing of arbitrary-resolution images. To understand the contribution of different training data categories, we conduct a leave-one-out data mixture ablation study-systematically removing task, domain, modality, and language categories-to diagnose which data types are necessary versus redundant and whether task benefits transfer across domains. Model weights and code are publicly released at https://huggingface.co/jinaai/jina-vlm. |
| title | jina-vlm: Small Multilingual Vision Language Model |
| topic | Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition 68T50 I.2.7 |
| url | https://arxiv.org/abs/2512.04032 |