Research Papers
CaSKG: Counterfactual-Causal Skill Graphs Boost LLM Agent Retrieval
A new framework from Hugging Face researchers calibrates skill relations using counterfactual-causal graphs, improving retrieval a...
CaRGo-T: Graph-of-Thought Framework Boosts Multimodal Humor Comprehension in VLMs
Researchers propose CaRGo-T, a causal reasoning graph-of-thought framework that improves humor understanding and detection in visi...
Hugging Face Unveils Luce: AI Model Generates Relightable 3D Assets from Single Images
Luce, a new 3D representation from Hugging Face, unifies geometry and PBR materials in a voxelized Gaussian cloud, enabling high-f...
Hugging Face Unveils Magpie: A Real-Time Generative Renderer for Interactive Games
Magpie, a new system from Hugging Face, separates game logic from visual generation to enable real-time generative rendering while...
CritICL: Turning Small Model Failures into Efficient Reasoning Guidance
Hugging Face researchers introduce CritICL, an inference-time framework that uses structured failure patterns from weaker models a...
Procedura: AI Agent That Writes 3D Models as Editable Code
Hugging Face researchers introduce Procedura, an agentic 3D modeling framework that generates editable, part-structured procedural...
Hugging Face Paper Proposes ACE Lens for Agentic Data Generation
A new paper on Hugging Face introduces a unified framework for agentic data generation, emphasizing accuracy, complexity, and dive...
Self-OPD: Teacher-Free On-Policy Distillation for Flow Matching Models
Hugging Face researchers introduce Self-OPD, a teacher-free on-policy distillation framework for flow matching models that uses se...
TTPO: Label-Free Test-Time Training Matches Supervised Math Reasoning
Hugging Face researchers introduce Test-Time Policy Optimization (TTPO), a label-free method that distills agreeing rollouts and p...
UrbanGround: New Sandbox Reveals Limits of AI Agents in Real-Scale City Navigation
Researchers introduce UrbanGround, a realistic 3D replica of Hong Kong, to test whether multimodal AI agents can sustain navigatio...
PAWBench: Probabilistic Alignment in World Models Remains Elusive
A new benchmark, PAWBench, reveals that current video generation models fail to match reference behavior distributions, highlighti...
Hugging Face Proposes RLHEV: Using Game Engines as Verifiable Data Engines for World Models
A new paper from Hugging Face argues that scaling world models requires more than just more video data and compute. It proposes Re...