useful sharded checkpoints for users to run inference / fine-tuning on a Google colab without having to deal with CPU OOM issues.