AI
Zyphra Redefines Efficiency in Vision-Language Models with Zamba2-VL
Zyphra introduces Zamba2-VL, an open-source family of hybrid Mamba2 state-space and Transformer vision-language models (1.2B, 2.7B, 7B parameters). Zamba2-VL significantly reduces 'time-to-first-token' by an order of magnitude while maintaining competitive performance against pure Transformer models, released under an Apache 2.0 license.
Carlos "Emérito" López Lovera
