textattack/roberta-base-STS-B

TextAttack Model Card

This roberta-base model was fine-tuned for sequence classification using TextAttack and the glue dataset loaded using the nlp library. The model was fine-tuned for 5 epochs with a batch size of 8, a learning rate of 2e-05, and a maximum sequence length of 128. Since this was a regression task, the model was trained with a mean squared error loss function. The best score the model achieved on this task was 0.9108696741479216, as measured by the eval set pearson correlation, found after 4 epochs.

For more information, check out TextAttack on Github.