BERT Extractive QA on SQuAD v1.1
Intended Use
English Extractive Question Answering using the SQuAD v1.1 dataset, developed for an academic course assignment.
Model and Training Details
- Base Model: bert-base-uncased
- Dataset: rajpurkar/squad (Subsampled to 15k training examples).
- Adaptation Strategy: Full Fine-tuning with differentiated learning rates; random seed 42; 2 epochs; batch size 16.
- Optimizer Configuration: Discriminative learning rates (Encoder LR: 2e-5, QA Head LR: 1e-3).
Evaluation Results
Evaluation performed on the subsampled validation set (2,000 examples):
- Exact Match (EM): 60.70
- F1-Score: 71.74
- Validation Loss: 1.319787
References
- Rajpurkar et al. (2016), SQuAD: 100,000+ Questions for Machine Comprehension of Text.
- Devlin et al. (2019), BERT: Pre-training of Deep Bidirectional Transformers.