Research Papers
Hugging Face Paper Introduces Causal Foundation Models for In-Context Causal Inference
A new paper from Hugging Face presents Causal Foundation Models (CFMs), pretrained networks that estimate causal effects on new da...
Hugging Face Unveils ENEAS: A Text-Promptable Method for Robust Instance Tracking and Semantic Discovery
ENEAS unifies text-prompted instance tracking and open-concept semantic discovery, addressing temporal hallucinations, spatial fra...
Hugging Face Researchers Unveil Uno: Lossless Diffusion Speedups for LLMs
A new class of diffusion-augmented LLMs, named Uno, enables parallel token sampling to accelerate inference by up to 3x without qu...
Hugging Face Researchers Introduce RevPropBench to Test LLMs' Revision Propagation in Conversational Artifacts
A new benchmark, RevPropBench, evaluates how well LLMs propagate local revisions across dependent elements in conversationally gen...
Hugging Face Unveils FlowBalance: Verifier-Grounded Self-Improvement for Reasoning Models
FlowBalance enhances reasoning models by calibrating self-guidance with verifier feedback, improving performance and stability ove...
Hugging Face Researchers Introduce Teacher-Gated Distillation to Boost AI Training Efficiency
A new method, Teacher-Gated On-Policy Distillation (TGOPD), verifies teacher reliability per prompt before applying dense supervis...
EditVid: A Unified Training-Free Framework for Diverse Video Editing
Hugging Face researchers introduce EditVid, a training-free framework that unifies instruction-guided and subject-guided video edi...
GAPO: Adaptive Clipping Boosts RLVR by Preserving Rare Success Signals
Hugging Face researchers propose Group Adaptive Clipping Policy Optimization (GAPO), a plug-in modification to GRPO that adapts cl...
Hugging Face Study: LLMs Over-Edit Code, But RL Can Teach Minimal Fixes
New research from Hugging Face shows that LLMs often over-edit code during repair, but preservation instructions and reinforcement...
Hugging Face Introduces τ^τ-Bench: A Realistic Test for Building Customer-Service Agents
A new benchmark from Hugging Face evaluates coding agents on constructing production-ready customer-service agents from real busin...
FactoSR: A Factorized Reinforcement Learning Framework Boosts 4D Spatial Reasoning in VLMs
Researchers introduce FactoSR, a reinforcement learning framework that decomposes 4D spatial reasoning into XY, Z, and T sub-objec...
MaxKernel: Multi-Agent System Automates Expert-Level TPU Kernel Generation
Hugging Face researchers introduce MaxKernel, a multi-agent system that automates TPU kernel development via three paradigms, achi...