My particular focus in Benchopt was the design of a deep learning benchmark. In its first iteration it features 2 frameworks, PyTorch and TensorFlow, 3 datasets, 3 CNNs and 4 optimizers. I am dedicated to growing it to a larger scale, i.e. with ImageNet, handling multiple GPUs (even maybe multiple nodes), including Jax. Another improvement direction is to build a benchmark that includes the hyperparameter tuning in the overall optimization.