Hugging Face Daily Papers

AETHER: a unified framework enabling geometry-aware reasoning in world models, achieving zero-shot generalization from synthetic to real-world data.

AlphaSpace enables robotic actions via semantic tokenization and symbolic reasoning, enhancing spatial intelligence in LLMs.

Hummingbird: An efficient text-to-video model that balances quality and computational efficiency via pruning and visual feedback learning.

CFG-Zero*: A better Classifier-Free Guidance to improve the image quality and text alignment in Flow Matching models.

Diffusion-4K: Synthesizing ultra-high-resolution images with a new benchmark dataset and wavelet-based fine-tuning that makes 4K image creation more detailed and accessible!

Aligning image generation subtasks: Equivariant modeling boosts efficiency and generalization by leveraging natural visual signal invariance.

FFN Fusion: Parallelizing sequential computation in large language models for significant speedups!

LLMs’ reasoning is decoded via sparse autoencoders, revealing key features that, when steered, enhance performance. First mechanistic account of reasoning in LLMs!

MetaSpatial: RL for 3D Spatial Reasoning in VLMs