Search
On October 30, the Beijing Academy of Artificial Intelligence (BAAI) introduced the Emu3.5 multimodal world model. This model was trained on an extensive dataset, encompassing over 10 trillion tokens of multimodal data. Such a robust training regimen has dramatically extended the model's video data training period, escalating it from an equivalent of 15 years to a staggering 790 years. Moreover, the model's parameter count has seen a substantial increase, rising from 8 billion (8B) to 34 billion (34B). When it comes to multimodal storytelling, Emu3.5 excels at crafting captivating narratives centered around any specified theme, thereby igniting an endless stream of imagination. Furthermore, the model is equipped with cross-scenario embodied operational abilities, facilitating generalized action planning and facilitating intricate interactions.
-
C114 Communication Network -
Communication Home
