batch_size = 64 steps = 10000 print_every = 500 seed = 1 cnn_storage = "./res/cnn.eqx" [[sae]] layer = 6 hidden_size = 256 input_size = 64 learning_rate = 1e-4 l1 = 3e-4