Gravitational lensing offers a powerful probe into the properties of dark matter and is crucial to infer cosmological parameters.The Legacy Survey of Space and Time (LSST) is predicted to find O(10^5) gravitational lenses over the next decade, demandingautomated classifiers. In this work, we introduce GraViT, a PyTorch pipeline for gravitational lens detection that leveragesextensive pretraining of state-of-the-art Vision Transformer (ViT) models and MLP-Mixer. We assess the impact of transferlearning on classification performance by examining data quality (source and sample size), model architecture (selectionand fine-tuning), training strategies (augmentation, normalization, and optimization), and ensemble predictions. This studyreproduces the experiments in a previous systematic comparison of neural networks and provides insights into the detectabilityof strong gravitational lenses on that common test sample. We fine-tune ten architectures using datasets from HOLISMOKESVI and SuGOHI X, and benchmark them against convolutional baselines, discussing complexity and inference-time analysis.Our publicly available fine-tuned models provide a scalable transfer learning solution for gravitational lens finding in LSST.

The .zip file contains the fine-tuned models for all the experiments in GraViT. There are 12 transfer learning settings and 10 architectures, resulting in 120 models.