Better use_one_hot_embeddings documentation in extract_features

[jonthegeek]

The documentation for use_one_hot_embeddings says: "Logical; whether to use one-hot word embeddings or tf.embedding_lookup() for the word embeddings."

I can get most of that from the name of the parameter. What's the difference between those two options?

Also update .model_fn_builder_EF to inherit this param from extract_features (or vice versa?), so it's documented the same in both places. Actually, BertModel also uses it, and it looks like its "home" is embedding_lookup. The documentation there is slightly different, but neither really helps me grok what the difference is/when I'd want it to be TRUE.

[jonathanbratt]

Yeah, I was just thinking about this the other day. I'm not even sure this should be a parameter in extract_features. Will investigate.