Model based on distilbert-base-uncased model trained on natural question short dataset. Trained for one episode with AdamW optimizer and learning rate of 5e-03 and no warmup steps. We achieved a f1 score of 32.67 and an em score of 10.35
Model based on distilbert-base-uncased model trained on natural question short dataset. Trained for one episode with AdamW optimizer and learning rate of 5e-03 and no warmup steps. We achieved a f1 score of 32.67 and an em score of 10.35