The switch-base-32 model was fine-tuned on the SQuAD dataset. Validation exact-match/F1-score: 65.39/85.81. The prompt key in PromptSource: "answer_given_context_and_question".