GPT-2 doesn't include dropout layers

[bwdGitHub]

We would like to use these issues to gauge user interest.

The GPT-2 implementation does not include dropout layers. This would be useful for further pre-training and fine-tuning workflows to prevent overfitting.