bert-base-uncased finetuned on MNLI

Model Details and Training Data

We used the pretrained model from bert-base-uncased and finetuned it on MultiNLI dataset.

The training parameters were kept the same as Devlin et al., 2019 (learning rate = 2e-5, training epochs = 3, max_sequence_len = 128 and batch_size = 32).

Evaluation Results

The evaluation results are mentioned in the table below.

Test Corpus Accuracy
Matched 0.8456
Mismatched 0.8484
Downloads last month
Hosted inference API
Text Classification