Generate ELMO from Glove about bilm-tf HOT 4 CLOSED

I think ELMO is contextualized. Same word in different sentences maps different vec in ELMO, whille Glove maps one fixed vec in all sentences. They are totally diffenrent.

from bilm-tf.

@zxgineng so if I need to replace Glove I have to use the dataset direct to generate a vector per token not to look in the glove file and find the equivalent vector correct?
In other words, in glove we use the data set tokens and check for each token to the equivalent word in glove file and take the corresponding vector if found.
Now instead of looking inside glove file, we are going to generate for each token an ELMO vector.
Correct? If yes before per token we have 300d vector in ELMO what's the dimensionality?

from bilm-tf.

"Now instead of looking inside glove file, we are going to generate for each token an ELMO vector.
Correct?" -- Yes, this is correct. You will compute an ELMo vector for each token in your input sentence.

"If yes before per token we have 300d vector in ELMO what's the dimensionality?" -- for the provided pre-trained model, it's 1024.

from bilm-tf.

for the provided pre-trained model, it's 1024.

So, the pretrained model is a plug-and-play one, just like GloVe? Or we need to train it too in the current context?

from bilm-tf.