X · @teortaxesTex
· X / Twitter
Bytedance continues to chase GDM "Reasoning" in pure image generation, trained with – you guessed it! – another variant of GRPO.
Bytedance continues to chase GDM"Reasoning" in pure image generation, trained with – you guessed it! – another variant of GRPO.DailyPapers: ByteDance just released UniVR-34B on Hugging FaceThe first model to learn complex reasoning, physical dynamics, and long-term planning directly from visual demonstrations — no text