Research Papers
LightNav-0: Compact VLM-Based Model Achieves SOTA Embodied Navigation
Hugging Face researchers introduce LightNav-0, a compact generalist navigation model that harnesses pretrained vision-language mod...
SHAPE Framework Decodes Chain-of-Thought Reasoning in LLMs for Better Math Performance
Researchers introduce SHAPE, a framework that analyzes chain-of-thought reasoning through semantic spaces and heuristics, revealin...
Qwen3.8-Flash-Next: A Sparse MoE That Slashes Compute While Boosting Stability
A new paper details Qwen3.8-Flash-Next, a 125B-parameter sparse mixture-of-experts model that activates only 6B parameters per tok...
NoRA: Normalizing LoRA Down-Projections for Faster, More Stable Training
Researchers introduce Normalized Low-Rank Adaptation (NoRA), a simple tweak to LoRA that normalizes down-projection matrices, acce...
Hugging Face Unveils CogEvol: Single-Pass Learning Environment Generation
CogEvol, a new family of models from Hugging Face, generates structured learning artifacts like slides and interactive HTML pages...
PaperGym: Turning Papers into Training Environments for AI Research Planning
Hugging Face researchers introduce PaperGym, a framework that converts scientific papers into reinforcement learning environments...
Study: On-Policy Distillation Works by Suppressing Low-Probability Tokens, Not Teacher Guidance
A new Hugging Face paper reveals that on-policy distillation (OPD) improves reasoning mainly by suppressing low-probability tokens...
Hugging Face Unveils Lucida: A New Pipeline for Composable Real-to-Sim Scene Modeling
Lucida, a new method from Hugging Face, improves composable indoor scene reconstruction by redistributing pipeline requirements ac...
GenFirst: A New Strategy for Stable End-to-End Latent Generative Modeling
Researchers introduce GenFirst, a generation-before-reconstruction strategy that enables stable direct end-to-end training of late...
Hugging Face Unveils DreamX-Creator: 7B Model for Native 2K Audio-Video Generation
DreamX-Creator 1.0 is a compact 7B model that natively generates synchronized audio and video at 2K resolution, using cross-modal...
New EASEL Benchmark Reveals Multimodal Agents Struggle with Dexterous Visual Tool Use
Hugging Face researchers introduce EASEL, a benchmark for fine-grained visual tool use, showing that current multimodal agents fai...
Sliding-Window Attention Outperforms Linear Attention on Long Contexts
A new study from Hugging Face shows that sliding-window attention with sinks matches or beats post-trained linear attention on lon...