xwei00/DRRho-CLIP-ViT-B-16-DFN-192M-Ref-ViT-B-32

Model Steering: Learning with a Reference Model Improves Generalization Bounds and Scaling Laws

Paper link: arXiv

Code & more: GitHub

This repository hosts a ViT-B/16 CLIP model trained using DRRho-CLIP with OpenAI ViT-B/32 CLIP model as reference. The training dataset is DFN-192M and the number of samples seen is 1.28B. Commands and code to reproduce are provided in the GitHub repository, and more details can be found in the paper.