Salamandra VL 7B 2512
Salamandra VL 7B 2512 updates BSC's visual-language model with a new image encoder and additional multilingual training.
Research release
Model and training
It combines Google SigLIP 2 Giant patch-16 with BSC's Salamandra 7B base and a two-layer projector. Training adds Aya language instructions and PixMo examples for visual grounding, counting and document understanding. The architecture follows LLaVA OneVision. Its research-only license differs from the Apache terms of the text-only Salamandra base.Salamandra VL 7B 2512: official model card
Practical use
Practical guidance. Use this release to study whether the same image can be discussed accurately in several European languages. Inspect object counts, document reading and references to image locations separately. A correct answer in one language does not establish equivalent performance in another. Check that the research-only terms match the intended use.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Salamandra-VL-7B-2512 |
| License | RESEARCH-ONLY RAIL-AMS.Salamandra VL 7B 2512: official model card |
| Model size | 7B language backbone plus visual encoder and projector.Salamandra VL 7B 2512: official model card |
Related models
- Salamandra 7B
Barcelona Supercomputing Center · Salamandra
Seven-billion-parameter multilingual base model that forms the language backbone of later Salamandra adaptations.
Launch date not confirmedDownloadable - Salamandra 7B Vision
Barcelona Supercomputing Center · Salamandra Vision
Earlier image and video understanding research model that has been superseded.
Launch date not confirmedResearch releaseDownloadable