learning_rate:

• 1e-5

train_batchsize:

• 16

epochs:

• 2

weight_decay

• 0.01

optimizer

• Adam

datasets:

• squad

metrics

• EM:10.307414104882
• F1:42.10389032370503