Closed Lguyogiro closed 11 months ago
This adds a decoder dropout option so that you can set different dropouts in the different task decoders if desired. The dropout is applied to the mlm embeddings as a first step in the decoder. The decoder dropouts default to zero.
Looks good!, would you mind to add a documentation page?, it could be in the list of "other things" in the main README?
Don't forget to acknowledge yourself at the bottom of the page
This adds a decoder dropout option so that you can set different dropouts in the different task decoders if desired. The dropout is applied to the mlm embeddings as a first step in the decoder. The decoder dropouts default to zero.