Latest stories
REDE: Denoising Reasoning Traces to Boost Hallucination Detection in Large Reasoning Models
Researchers propose REDE, a framework that denoises reasoning traces by removing irrelevant and repetitive steps, improving halluc...
Frozen 12B Model Achieves 100% Accuracy via Verified Memory, Not Raw Reasoning
A new approach from Hugging Face stores verified solutions in a persistent memory, enabling a frozen 12B model to answer 180/180 i...
Chamaileon: A Unified Framework for Multi-Target and Multi-State Protein Binder Design
Researchers introduce Chamaileon, a framework that unifies multi-target and multi-state protein binder design using in-context co-...
ClinFusion: Vision-Centric MLLM Sets New State-of-the-Art in Medical Understanding
Alibaba DAMO Academy introduces ClinFusion, a vision-centric multimodal LLM that unifies 2D and 3D medical image understanding, ou...
Hugging Face Researchers Use RAG to Restore Historical Documents
A new framework called ARI combines large language models with retrieval-augmented generation to restore illegible historical docu...
dRAE: Hyper-Spherical Quantization Solves Codebook Collapse in Visual Tokenization
Hugging Face researchers propose dRAE, a visual tokenizer using Hyper-Spherical Quantization (HSQ) to achieve 100% codebook utiliz...
Hugging Face Paper Unravels the Physics of Multi-Turn Long-Horizon Planning for AI Agents
A new research paper introduces a controlled environment to systematically study how foundation model agents acquire, shape, and i...
Hugging Face Unveils Oxygen-TryOn: Open-Source Foundation Model for Any-Item Virtual Try-On
Oxygen-TryOn is a fashion-native foundation model that enables photorealistic virtual try-on of any item—from clothing to accessor...
OmniVAE: Jointly Trained Audio-Video VAE Achieves Cross-Modal Alignment for Synchronized Generation
Hugging Face researchers introduce OmniVAE, a jointly trained audio-video VAE that uses contrastive learning and semantic distilla...
Sol-Attn: Training-Free Sparse Attention Speeds Up Video Generation by 2x+
Sol-Attn introduces a training-free dynamic sparse attention method that unifies routing, computation, and approximation in a sing...
Hugging Face Researchers Expose Failure Mode in Diffusion Distillation with Classifier-Free Guidance
A new study reveals that naive matching of guided predictions in on-policy diffusion distillation can lead to 'Negative Branch Asy...
Hugging Face Researchers Propose 'Data Pyramid' to Organize Embodied AI Data Sources
A new paper from Hugging Face introduces the Embodied Data Pyramid, a taxonomy organizing five complementary data sources for embo...