learning rate: 5e-5   
training epochs: 5   
batch size: 8   
seed: 42  
model: bert-base-uncased  
trained on CB which is converted into two-way nli classification (predict entailment or not-entailment class)