How can I get the required HDF5 file

Hi, I’m a newbie to NLP field.
I am currently trying to play with ELMO with pytorch in AllenNLP, and it seems it requires json and HDF5 files to work. From what I understand, they were generated by training bilm(I am trying to train Chinese so I couldn’t just use the HDF5 given in the tutorial right?).
I feel I could just this package to first train a model https://github.com/allenai/bilm-tf, but it says:" We also have a pytorch implementation available in AllenNLP.", that would be really nice because I dont have to use tensorflow or keras for pre-training, but I did not really find how to do so in AllenNLP. Could anyone shade some light on it?
Thanks a lot!

@howie do you want to retrain ELMo or do you want to use the pretrained model? Our pretrained ELMo model was trained with the bilm repository in TensorFlow, but there is also a transformer model that was trained within AllenNLP (see https://allennlp.org/elmo).

Hi, thanks for the reply.
I wish to train a new model from the start since I am trying it on Chinese NLU, hence the question about how to get HDF5 file using allennlp. If there is a pre-trained model for this task, it would be nice. Still, I wish to know how to generate pre-train a model myself as if there is no pre-train model given(I know this could be done by bilm-tf, but what about allennlp with pytorch? Since in bilm-tf it says:" We also have a pytorch implementation available in AllenNLP." Not sure if it was included in Allennlp)

@howie, see https://github.com/allenai/allennlp/blob/master/tutorials/how_to/training_transformer_elmo.md for the Transformer-based version implemented in PyTorch. I should note, however, that we haven’t worked on making ELMo multi-lingual, so you’ll likely need to make some non-trivial modifications to have it run on Chinese text.