Re: [ot][spam][crazy][data] transformer model 'attention' improvement

[k]

i found the 'reformer' at
https://huggingface.co/transformers/v2.9.1/model_doc/reformer.html .
this architecture appears to be about 2 years old. it says it is
comparable to a transformer, but much more memory efficient, and uses
a different kind of 'attention'.