Zamba2-VL
#51 v Multimodální modelyZyphra · v2 · vl · 3× · naposledy 29. 6. 2026
3
Momentum
Zamba2-VL is a language model by Zyphra released under the Apache 2.0 license. The model extends hybrid architectural ideas that combine SSM (State Space Models) and Transformer approaches.
Vývoj momenta
19.05.17.08.
Vlastnosti
| Key Benchmark (%) | PixMoCount: 62.5 (1.2B model); DocVQA: 90.9 (2.7B model) |
| License | Apache 2.0 |
| Multimodality | Vision-language: single- and multi-image understanding, visual grounding, OCR, document/chart understanding, visual counting |
| Platform | Open weights on Hugging Face (Zyphra/Zamba2-VL-1.2B/2.7B/7B), inference code on GitHub |
| Release Date | June 10, 2026 (announcement), preprint May 29, 2026 |