This work jointly varies model size, training tokens, and forward and backward loop counts across nearly 600 training runs and fit a unified scaling law that captures the effects of looping and truncation through effective parameters, showing that at fixed model size, the compute-optimal loop count and training-token b...
Xu-Yuan Xiong, Si-Yu Chen, Zi-Han Shao et al.· 1 citation
Autoregressive video generation requires denoising the current frames while writing their key-value representations as context for future predictions. However, these two roles typically share parameters, and we find that their gradients exhibit distinct patterns and systematic negative alignment, hindering the joint op...
Zi-Han Su, Junhao Zhuang, Yao-Wei Li et al.· 3 citations
EchoWM is an omnimodal world model for enterable generative media that responds to continuous navigation while jointly generating 720p video, environmental sound, music and speech, and maintains synchronized environmental sound and speech over long-horizon generation.
Self Gradient Forcing (SGF), a two-pass training strategy that restores this missing memory-writing supervision within the native autoregressive training objective, using losses on future video latents to train the model to encode context into more effective causal memory.
Results indicate that memory, geometric control, and rollout-aware training provide a practical foundation for generating coherent stories and continuously evolving interactive worlds.
Nan Duan, Hao-Yang Huang, Wei-Yang Jin et al.· 4 citations· ⚡1
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
