Can I train a language model and generate weights file by myself？ about bilm-tf HOT 5 CLOSED

Sorry, no we haven't released the training code.

from bilm-tf.

In order to create ELMO embeddings for my own dataset, do I need to use the trained weights provided by the following link : https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5

Thank you

from bilm-tf.

Yes, that is the weight file for the pre-trained biLM. You will need the options file that specifies the hyperparameters. These both linked from README.md.

from bilm-tf.

Thank you.
That is the weight file for the pre-trained biLM. I will use it with its options file like you mentioned in README.md. So in order to create ELMO embeddings for questions and paragraphs separately, which of your example python files I need to follow? (usage_cached.py or usage_token.py) .
Can you please correct me if I am wrong? Per my understanding, usage_token is like Glove weights right? It provides weights for each token in voc?

from bilm-tf.

Can you provide more detailed instructions to train my own elmo embedding in your README? I want to train elmo on NYT and compare against word2vec or glove.

from bilm-tf.