Research Papers
LLM Reasoning Operations Map to Separable Geometric Structures in Hidden States
A new study from Hugging Face researchers reveals that distinct reasoning operations in LLMs, such as deduction and goal decomposi...
UniMate: One Diffusion Model to Animate Any Skeleton from Text and 3D Assets
Hugging Face researchers introduce UniMate, a unified diffusion transformer that generates articulated motion for arbitrary skelet...
Hugging Face Unveils Enoki: A Unified Framework for Efficient Multi-Level Hallucination Detection
Enoki, a new open information extraction framework from Hugging Face, unifies claim-level verification and span-level hallucinatio...
Attention Triangle Reveals Bidirectional Semantic Leakage in Audio-Video Diffusion Models
Researchers analyze the 'attention triangle' in audio-video diffusion models, showing that cross-modal attention causes bidirectio...
Dr. Claw: An Open-Source Workspace for Auditable AI Research Workflows
Hugging Face researchers introduce Dr. Claw, an open-source workspace that wraps existing command-line coding agents in a controll...
Hugging Face Paper Formalizes Multi-Agent LLM Coordination with Game Theory and New SRMA Algorithm
A new paper on Hugging Face models orchestrator-worker LLM systems as bilevel coordination games, proving limits of transcript-onl...
Hugging Face Researchers Unveil RISE: A Self-Extrapolating Method to Boost AI Reasoning
A new paper from Hugging Face introduces RISE, a training method that uses a model's own reinforcement learning trajectory to gene...
Reviving Layer Dropout: New Study Shows Efficiency Gains in LLM Training and Inference
A comprehensive study from Hugging Face demonstrates that layer dropout, when properly configured, can reduce training FLOPs by up...
Hugging Face Researchers Introduce OR-Clarify and InterOPT to Make AI Ask Before Optimizing
A new benchmark and framework from Hugging Face researchers teach LLMs to ask clarifying questions before formulating optimization...
Motion-Omni: End-to-End Framework Unifies Speech and Full-Body Motion for Spoken Dialogue
Hugging Face researchers introduce Motion-Omni, an end-to-end framework that jointly generates spoken dialogue and full-body co-sp...
WorldSculpt: Turning Cluttered Videos into Compositional 3D Worlds
Hugging Face researchers introduce WorldSculpt, a method that adapts a single-object 3D generative prior to multi-view video obser...
Hugging Face Unveils Iris: Open-Source Search Agents Climb to Frontier Performance
Hugging Face presents Iris-mini and Iris-pro, two open-source search agents trained via a novel SFT-RL climbing pipeline, achievin...