#GSPO18/11/2025
Uni-MoE-2.0-Omni: An Open Omnimodal MoE Unifying Text, Image, Audio and Video
'Uni-MoE-2.0-Omni is an open omnimodal MoE on a Qwen2.5-7B core that unifies text, image, audio and video via a shared token interface, dynamic expert routing and staged training for robust multimodal understanding and generation.'