Model card for detoxified gpt-j-6b

Model run can be found here

The main difference is that I used mini_batch_size=1