Read the original at HF Daily Papers
LARC enables unified multimodal models to perform latent reasoning in continuous space, improving text-to-image generation through curriculum supervised fine-tuning and information-gain reinforcement learning.
Carried by: HF Daily Papers. First seen: .