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.