Research Papers
Self Gradient Forcing: New Method Boosts Long Video Generation from Short Training Clips
Researchers propose Self Gradient Forcing (SGF), a two-pass training strategy that enables autoregressive video diffusion models t...
SLAI T-Rex Achieves 34.22% MFU on Ascend NPU for DeepSeek-V4 Post-Training
A new hierarchical optimization framework enables full-parameter post-training of trillion-parameter MoE models on Ascend NPU Supe...
Hugging Face Unveils AlayaWorld: Open-Source Interactive Video World Model
AlayaWorld is an interactive long-horizon video world model that generates 24-fps video at 540p and 720p from text, image, or vide...
Hugging Face Unveils Mage-Flow: A Compact 4B Model for Efficient Image Generation and Editing
Mage-Flow is a 4B-parameter generative stack that achieves competitive quality in text-to-image generation and instruction-based e...
Hugging Face Study Reveals 'Useless' Text Tokens Are Implicit Semantic Registers in Diffusion Transformers
A new causal interpretability framework shows that structural template tokens in text-to-image diffusion transformers, though carr...
AlayaRenderer-Flash: Real-Time Generative World Rendering Hits 30 FPS
Hugging Face researchers unveil AlayaRenderer-Flash, a real-time generative world renderer that accelerates frame synthesis from 0...
DataFlow-Harness Bridges the NL2Pipeline Gap with Grounded Code Agents
Hugging Face researchers introduce DataFlow-Harness, a platform that uses LLM agents to construct editable DAG-based data pipeline...
ABot-World-0: Real-Time Interactive World Model Runs at 16 FPS on a Single RTX 5090
Hugging Face researchers introduce ABot-World-0, an action-conditioned video world model that enables real-time, long-horizon clos...
Hugging Face Unveils Open-AoE: 2,000-Hour Egocentric Manipulation Dataset for Embodied AI
Open-AoE is an open-source dataset and toolchain for egocentric manipulation, featuring 2,000 hours of video from 500+ contributor...
SWE-Pruner Pro: Pruning Code Context Using the Agent’s Own Internal Signals
Hugging Face researchers introduce SWE-Pruner Pro, a method that prunes tool outputs in coding agents by leveraging the agent’s in...
DeepSearch-Evolve: Self-Distillation Enables Scalable Self-Improvement for Web Agents
Hugging Face introduces DeepSearch-Evolve, a self-distillation framework that trains web agents to improve from their own experien...
EvolvingWorld: Open-Schema Framework for Co-Evolving Characters and Worlds in Literary Simulation
Hugging Face researchers introduce EvolvingWorld, a framework and benchmark for long-horizon literary world simulation where chara...