textattack/albert-base-v2-STS-B

TextAttack Model Card

This albert-base-v2 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 32, a learning rate of 3e-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 0.9064220351504577, as measured by the eval set pearson correlation, found after 3 epochs.

For more information, check out TextAttack on Github.