decoder_stack.TransformerTaskConfig:
  sequence_length = 512
  batch_size = 8

transformer_layer.TransformerLayer:
  window_length = 512
  use_long_xl_architecture = False