Research Papers
Hugging Face Paper Unveils GroundAnything, a 4B Diffusion Model for Flash-Speed Visual Grounding
GroundAnything reframes visual grounding as parallel visual evidence extraction, using blockwise denoising to deliver precise loca...
SpatialASM Teaches Vision-Language Models to Mentally Re-Imagine Scenes
Hugging Face researchers propose SpatialASM, a training framework that internalizes Active Spatial Mentalization in large vision-l...
AgentSWE Benchmark Tests Whether Coding Agents Really Build the Agent You Asked For
Hugging Face researchers introduce AgentSWE, a 25-task benchmark spanning agent creation, codebase editing, and optimization, find...
Multi-Stream Attention Residuals Boosts Large-Scale Pre-Training Efficiency
A new paper introduces Multi-Stream Attention Residuals (MSAR), a residual architecture that generalizes depth-axis attention to m...
Hugging Face Paper Unveils GroundingPI, a 4B Grounding Foundation Model for Physical Intelligence
GroundingPI is a 4B-parameter grounding foundation model that outputs points and boxes as quantized coordinates in a shared vocabu...
Hugging Face Paper Introduces TerminalHorizon: Scaling Long-Horizon Agentic Data for Terminal Intelligence
A new research paper presents TerminalHorizon, an automated data synthesis engine that builds long-horizon terminal agent trajecto...
InstantFusion Bridges Heterogeneous Image Generators via a Shared Latent Interface
A new Hugging Face paper introduces InstantFusion, a lightweight shared latent interface that lets independently trained diffusion...
DCOCO Restores Context to Parallel Drafts, Speeding Up Speculative Decoding
A new Hugging Face paper introduces DCOCO, a lightweight corrector that recovers intra-block dependencies lost by diffusion-based...
Hugging Face Paper: Representation Guidance Lets Plain Transformers Generate Pixels Without a VAE
A new paper introduces Representation Guidance (RG), a method that forms and exploits internal representations inside a plain pixe...
Hugging Face Paper Unveils PolyOCR-Venus, a Unified OCR Foundation Model Family
PolyOCR-Venus unifies text recognition, grounding, document parsing, information extraction, and OCR reasoning in a single instruc...
RSI-Anything Uses World Models to Drive Recursive Self-Improvement Across Domains
A new Hugging Face paper introduces RSI-Anything, a domain-general framework built on JEPA-Anything that leverages predictive worl...
Hugging Face Paper Tests Whether AI Agents Can Learn Their Way to the Top in Game Competitions
A new Hugging Face paper formalizes Adversarial Heuristic Learning (AHL), where AI agents refine executable game policies while mo...