This model is a fine-tuned SpanBERT base for extractive question answering on the SQuAD v1.1 dataset.
It was trained using the Hugging Face Trainer API with the following key hyperparameters:
Learning rate: 3e-5
Batch size: 8
Warmup ratio: 0.1
Weight decay: 0.01
Epochs: 2
Evaluation Results (SQuAD v1.1 Dev)
Using v1.1 dev dataset with official evaluation script, this is not uploaded to test on hidden testset.
