NASBench-101

The NASBench-101 dataset is a large-scale benchmark for neural architecture search. It contains nearly 423K unique convolutional neural network architectures with diverse structures and various number of convolutional operations.