NeoMME: an efficient Multimodal-native and Multilingual Encoder

We introduce NeoMME, a family of 260M and 800M multilingual multimodal encoders. Unlike many generative visual language models, NeoMME does not use a separate pretrained vision tower or a causal language model.