Zhangyang "Atlas" Wang

7 papers University of Texas at Austin

• Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights
• Graph-KV: Breaking Sequence via Injecting Structural Biases into Large Language Models
• Martian World Model: Controllable Video Synthesis with Physically Accurate 3D Reconstructions
• One Token Embedding Is Enough to Deadlock Your Large Reasoning Model
• REPA Works Until It Doesn’t: Early-Stopped, Holistic Alignment Supercharges Diffusion Training
• SAS: Simulated Attention Score
• Scaling Up Parameter Generation: A Recurrent Diffusion Approach