SimCLRv2 provides pre-trained models and Jupyter Notebook implementations for big self-supervised models that serve as strong semi-supervised learners. The framework trains deep neural networks (ResNet) to learn visual representations by maximizing agreement between differently augmented views of the same image. It includes TF2 implementations, colab notebooks for fine-tuning, and a collection of 65 pre-trained checkpoints across various model depths and widths.