Accuracies on the examples

[sf-wind]

Hi, I wonder whether you can provide some reference accuracy for the experiments you have put on the page: https://github.com/pbloem/former/tree/master/experiments

That helps me understand how well this model is.

Thanks

[TheGrayFrost]

I think he mentions in the blog that for sentiment classification at depth 6 and max sequence length 512, it achieves 85% classification accuracy. And for the text generation task, the model achieves a compression of 1.343 bits per byte on the validation set. You can obviously collect more metrics by yourself.
Hope it helps. Cheers! 👍