KyutaiCASA
CASA Qwen 2.5 VL 3B Shared
The shared variant ties self-attention and cross-attention weights in the Qwen-based adaptation.
Research release
The architecture determines how images enter the model
CASA lets text attend to visual features without inserting every image token into the language sequence. The code supplies a custom architecture and processor; this is not an unchanged upstream Qwen or Helium checkpoint.CASA-Qwen2_5-VL-3B-Shared: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/CASA-Qwen2_5-VL-3B-Shared |
| Weight license | CC BY-NC-SA 4.0.CASA-Qwen2_5-VL-3B-Shared: Kyutai model card |
| Upstream components | Qwen 2.5 VL 3B Instruct; its Qwen Research License also applies.CASA-Qwen2_5-VL-3B-Shared: Kyutai model card |
Related models
- Helium 1 VL 2B
Kyutai ยท CASA
An instruction-tuned vision-language reference model that inserts image tokens into the text sequence.
Launch date not confirmedResearch releaseDownloadable