Broken Neural Scaling Laws
Presents new Scaling Law that yields SotA extrapolation for each task within large, diverse set of downstream tasks, including large-scale Vision, NLP, Diffusion Models, 'Emergent' 'Unpredictable' Math, Double Descent, & RL.
This research paper can help businesses in optimizing their AI models by accurately modeling and extrapolating the scaling behaviors of deep neural networks for various architectures and for each of various tasks.
DIFFUSIONDB: A Large-scale Prompt Gallery Dataset for Text-to-Image Generative Models
Releases the dataset consisting of 2M images generated by Stable Diffusion using prompts and hyperparams specified by real users.
This dataset can aid researchers in understanding the interplay between prompts and generative models and could be used in designing human-AI interaction tools to help users more easily use these models.