Visualize: Figure 2 & 4.
Description: ": Inputs and outputs for the T5 Multi-task framework. In the reasoning mode, explanations are not conditioned on the model’s prediction, whereas in the rationalizing mode they are dependent on the model output"