Research Papers
SepGen Unifies Multi-Stem Audio-Video Separation and Generation in One Model
A new research paper introduces SepGen, which extends a pretrained audio-video generator to produce video, a mixed soundtrack, and...
Work2Skill Turns Human GUI Demos Into Reusable Skills for Computer-Use Agents
Hugging Face paper introduces Work2Skill, a framework that records human desktop GUI demonstrations and induces reusable agent ski...
SoftwareGym Synthesizes 182 Apps to Train Computer-Use Agents
Hugging Face researchers introduce SoftwareGym, an automated pipeline that synthesizes runnable GUI software environments to train...
Reflex Teaches Humanoid Robots to Catch Thrown Boxes Using Only Egocentric Vision
A new three-stage framework from Hugging Face researchers enables humanoid robots to reactively catch thrown boxes using whole-bod...
DiffusionART Cuts Diffusion RL Training Costs by Up to 12x, Researchers Say
A new paper introduces DiffusionART, a diffusion reinforcement learning method that removes hidden forward-reverse consistency for...
Hugging Face Paper Reveals 'S-Space,' an Internal Spatial Workspace Inside Multimodal AI Models
A new paper on Hugging Face identifies S-Space, an intrinsic linear subspace where multimodal language models encode object coordi...
SynthSelf Enables Continual Learning for Small Language Models
A new Hugging Face paper introduces Synthetic Self-Distillation (SynthSelf), a method that lets small language models acquire new...
SoloQ Brings Calibration-Free 4-Bit Quantization to Diffusion Language Models
A new Hugging Face paper introduces SoloQ, a calibration-free quantization framework for diffusion large language models that pres...
EmbodiedCLBench: New Benchmark Tests Whether Embodied Agents Can Truly Learn Continually
Researchers introduce EmbodiedCLBench, described as the first benchmark in embodied simulation for evaluating continual learning i...
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++,...