5fa1a76
1
2
3
The abstract from the paper is the following: We re-evaluate the standard practice of sharing weights between input and output embeddings in state-of-the-art pre-trained language models.
