Research Papers
IntBMoE Decouples Expert Participation, Compute, and Memory in Mixture-of-Experts
A new block-conditioned Mixture-of-Experts architecture, IntBMoE, achieves full expert participation while keeping execution spars...
OmniVBench and Omni-R2V Dataset Target Gaps in Reference-to-Video Generation
Hugging Face researchers introduce OmniVBench, a benchmark spanning 7 task families and 18 fine-grained tasks with 12,172 checklis...
Paint-Anything Brings Exact 24-Bit Hex Color Control to Image Generation and Editing
Hugging Face researchers introduce Paint-Anything, a unified hex-prompt interface that lets users specify an object's exact 24-bit...
Hugging Face Paper Unveils Code2Skill: Mining 1M Verifiable Agent Skills from GitHub Code
Code2Skill automatically extracts verifiable procedural skills from source code, producing a bank of over one million records that...
CodeMidas Turns Raw Source Code Into Scalable RL Training Environments for Coding Agents
A new Hugging Face paper introduces CodeMidas, an agentic pipeline that builds executable reinforcement-learning environments dire...
Hugging Face Paper Unveils RecreationWorld for Hybrid Computer-Use Agents
A new Hugging Face paper introduces RecreationWorld, a five-platform framework that trains hybrid computer-use agents to recreate...
EvoOntology Introduces a Self-Evolving Ontology Layer to Close the Agent-Data Gap
A new Hugging Face paper proposes EvoOntology, an MCP-based ontology layer that lets data agents query heterogeneous data at runti...
Qwen Open-Sources Qwen-Image-2.1, a 7B Unified Image Generation and Editing Model
Alibaba's Qwen team has released Qwen-Image-2.1, a compact open-source model that unifies text-to-image generation and image editi...
Hugging Face Paper Unveils UFO: A Unified Framework for Evaluating Multi-Modal Image Generation
UFO introduces a chain-of-evaluation paradigm that decomposes omni-condition alignment into atomic units, achieving a 15.25% avera...
Study Asks What Privileged Answers Really Add to On-Policy Self-Distillation
A new Hugging Face paper introduces AMPLE-Math, a 5,319-problem suite with six reasoning views, to isolate how much a teacher's ac...
When2Think: Difficulty-Aware Length Control Cuts Reasoning Token Use
A new post-training framework, When2Think, dynamically allocates reasoning computation based on problem difficulty, boosting AIME2...
FAMOS Brings Feed-Forward 3D Articulation Modeling to Sparse Views
Hugging Face researchers introduce FAMOS, a feed-forward model that predicts movable-part segmentation and joint parameters from a...