Efficient Large-Scale Visual Representation Learning And Evaluation