LLM-jp / NII LLMCJagle VL
Jagle VL 2.2B Jagle
Jagle VL 2.2B Jagle combines Qwen3-1.7B with SigLIP 2 So400m and a two-layer projector. It is an LLM-jp-trained multimodal derivative.
Research release
The dataset defines this release
The training corpus is Jagle, LLM-jp's Japanese multimodal dataset. The card declares Apache 2.0.Jagle VL 2.2B Jagle official model card
Compare the dataset choice at fixed architecture
Practical guidance. This Japanese-data variant is a useful control against FineVision-only training. Use Japanese images and questions alongside English tasks to examine which changes come from the dataset.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Jagle-VL-2.2B-Jagle |
| Language ancestry | Qwen3-1.7B.Jagle VL 2.2B Jagle official model card |
| Vision ancestry | Google SigLIP 2 So400m, patch16 at 512 pixels.Jagle VL 2.2B Jagle official model card |
| Training dataset | Jagle.Jagle VL 2.2B Jagle official model card |
Related models
- Jagle VL 2.2B FineVision
LLM-jp / NII LLMC · Jagle VL
The 2.2B Jagle vision-language experiment trained with FineVision data.
Launch date not confirmedResearch releaseDownloadable - LLM-jp-4 VL 9B
LLM-jp / NII LLMC · LLM-jp-4 VL
A Japanese-English vision-language reasoning model that changes both the language backbone and training data from the beta.
Launch date not confirmedResearch releaseDownloadable