koaning opened this issue 4 years ago · 1 comments
We could add one layer of self-attention to all the vocabulary-based models. That way there'd be context possible and we could have a unified API for [bank] of the river.
[bank] of the river
This idea totally does not work.