google-research / electra

ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators
Apache License 2.0
2.31k stars 351 forks source link

Improve Description of `--blanks-separate-docs`. #84

Open PhilipMay opened 3 years ago

PhilipMay commented 3 years ago

Hi. I do not fully understand the meaning of --blanks-separate-docs. I know that this is important for "next sentence prediction" with normal BERT. But I do not think that this is done at electra training. So my question is: How does this setting influence the training and the quality of the model? Or does it even influence the result at all?

Since I think that some other ppl. have the same question maybe this can also be a feature request for more documentation about this.

Many thanks Philip