How can I get the words embeddings? about xlm HOT 11 CLOSED

Hi!

Yes, I would suggest looking at the code in the notebook:
https://github.com/facebookresearch/XLM/blob/master/generate-embeddings.ipynb

Then doing something like that should work:

word_id = dico.index('cat')
model.embeddings.weight[word_id]

from xlm.

Got it, thx for everything!

from xlm.

Thank you! How can I differentiate words from different languages (ie 'chat' in french means cat)?

from xlm.

You can just replace "cat" by "chat" in the code above. There is only one share vocabulary, that contains the words for all languages. The vocabulary doesn't keep track of which word is used in which language.

from xlm.

i want get models for russian language, the mlm_xnli15_1024.pth will do?

from xlm.

Yes, it contains Russian. But these two models will give you a better performance, and also support Russian:

https://dl.fbaipublicfiles.com/XLM/mlm_17_1280.pth
https://dl.fbaipublicfiles.com/XLM/mlm_100_1280.pth

from xlm.

plus, for the bpe i should use this tokenization for mlm_17_1280.pth?

from xlm.

so, when i run to_bpe(sentences) i get next error
FileNotFoundError: [Errno 2] No such file or directory: '/tmp/sentences.bpe'

tmp folder i should create my hands? and where i should take sentences.bpe?

sorry, but your notebook have many errors at path to another users

from xlm.

also, what does it mean?
/private/home/aconneau/projects/XLM/data/wiki/17/175k/ i think this path are not available

from xlm.

plus, at folder tools i didn't give fastBPE
i should install from here. Is it's true?

from xlm.

solved errors and create new pr describing the steps

from xlm.