ATH MaaS (Alibaba)

An update to Ovis, the VLM series based on Qwen by another Alibaba subdivision.

Another omni model by another subdivision of Alibaba.

A VLM from Alibaba's AIDC Group, which learns and uses an image embedding lookup table, similar to the text part of other VLMs. The tokens from the encoder then can represent multiple of the words from the embedding table.

Another slightly different late fusion architecture for visual language models.