Research Papers
Hugging Face Paper Maps How Reasoning Transfers Across Model Scales via On-Policy Distillation
A new Hugging Face paper studies the scaling properties of on-policy distillation (OPD), finding a regular "useful-transfer" regim...
Survey Maps In-Context Learning for Robots From Demonstration to Execution
A new 100-page survey on Hugging Face organizes in-context learning for robots into four interface families and argues that evalua...
VoxMem Benchmark Finds Audio Language Models Fail at Spoken Memory
A new benchmark from Hugging Face researchers tests whether large audio language models can remember not just what was said, but w...
Hugging Face Paper Proposes 'Thinking Reward Model' That Scores Visual Generation Case by Case
A new paper introduces Think Before You Score, a paradigm in which a Thinking Reward Model builds case-adaptive rubrics before jud...
Hugging Face Paper Introduces Grounded Entity Biographies for Long-Video Memory
A new framework called Grounded Entity Biographies (GEB) links visually grounded observations of the same physical object across l...
Study Finds Future Modeling Is Key to World Action Model Generalization
A new empirical study shows that latent world action models lose the generalization benefits of future modeling, and proposes Simp...
Hugging Face Paper Unveils Raven, a 'Harness of Harnesses' for Composable AI Agents
Raven is an open-source multi-agent ecosystem that automatically builds and evolves modular harnesses for specific models and doma...
MaLiang-Harness Targets the 'Program-to-Visual Gap' in MLLM Image and Video Generation
A new Hugging Face paper introduces MaLiang-Harness, a framework that turns MLLM-driven visual generation into a persistent cycle...
QwenGyre: Elastic RL Framework Targets Ultra-Long-Horizon AI Agents
A new framework, QwenGyre, tackles GPU idling and trajectory redundancy in online reinforcement learning for agents whose single r...
Hugging Face Paper: Recursive Harness Distillation Boosts Robot Manipulation Success
A new research paper proposes Recursive Harness Distillation, a method that lets a strong agent distill intervention experience in...
FlyBy Teaches Small Reasoning Models When to Stop Thinking and Start Asking
A new Hugging Face paper distinguishes execution bottlenecks from knowledge bottlenecks in small reasoning models and introduces F...
Hugging Face Paper: UMM-Reflection Teaches Unified Models to Self-Repair Images
A new method applies interleaved reinforcement learning to full reflection trajectories inside a single unified multimodal model,...