Latest stories
VisionWeave Lets Multimodal Models Weave Visual Tokens on Demand
Hugging Face paper VisionWeave trains multimodal LLMs to allocate visual representations at content-adaptive granularity, saving 4...
StereoBind Ties Sound to Motion for Immersive Video-Audio Generation
A new Hugging Face paper introduces StereoBind, a framework that binds visual source motion to stereo sound generation, alongside...
Hugging Face Paper Unveils ProgramGym to Train Long-Horizon Coding Agents
ProgramGym introduces Program-Env and Program-Agent, a framework that builds 1,648 verifiable training environments across C, C++,...
Hugging Face Paper Proposes Incremental Deep Research That Updates Reports Instead of Rebuilding Them
A new Hugging Face paper introduces Incremental Open-Ended Deep Research, a setting where reports evolve as research states, suppo...
MIRA Agent Refines Text-to-Music Prompts to Match User Intent
Hugging Face researchers introduce MuRA-Bench, an expert-curated benchmark of 100 real-world music requests, and MIRA, a test-time...
SAP's STS Framework Generates Enterprise Data Without Database Schemas
A new Hugging Face paper introduces Synthesis Through Simulation (STS), a schema-free paradigm where an LLM agent generates enterp...
InfiLoop Adds Loop-Native Residuals to Stop Looped Transformers From Forgetting
A new paper argues that looped Transformers degrade as iterations grow because each loop overwrites the previous state, and propos...
SpaceFlow Brings Local Control to Text-to-3D Generation
A new training-free pipeline from researchers lets users assign per-region control levels to geometric primitives, balancing stric...
Hugging Face Paper Exposes 'Chaos in the Text' in Mixed-Modality Retrievers
A new Hugging Face paper finds that dense retrievers degrade sharply when text, image, and fused documents coexist, with irrelevan...
Hugging Face Paper: GPD Distills 3D Geometry Cues to Sharpen Spatial Reasoning in VLMs
A new paper introduces GPD, a geometry-privileged distillation method that routes question-relevant depth, semantic, and bird's-ey...
Hugging Face Paper: High Accuracy Doesn't Mean LLM Agents Admit Uncertainty
A new Hugging Face paper proposes evaluating LLM agents on "epistemic humility" — whether they recognize, act on, and communicate...
BrickBench Tests Whether AI Agents Can Actually Build With LEGO
Hugging Face researchers introduce BrickBench, a benchmark that asks coding agents to design buildable LEGO assemblies from text p...