{ "num_channels": 1, "input_sizes": [ 28, 28 ], "output_size": 10, "patch_size": 4, "num_heads": 28, "latent_size_multiplier": 15, "num_layers": 9, "encoder_size": 615, "head_size": 88, "dropout": 0.06, "encoder_activation": "gelu", "head_activation": "gelu" }