Github Reporsitory for Arxiv paper titled "Subword ELMo"
Pre-trained model can be downloaded from: https://pan.baidu.com/s/1cVVwNvvAuV_gy7uekMLBBA
If you want to use the pretrained model to enhance the representation of your text, you can use the function raw2embedding in file api.py.
If you want to train your own subword ELMo, first, you should change the configuration in config.py and then train the model by sh run.sh.(The training process will take too much time, we don't suggest anyone to waste the time and resource.)
python == 3.6.8
pytorch == 1.0.0
typing == 3.7.4.1
numpy == 1.15.0
progressbar == 2.5
overrides == 1.9
sentencepiece == 0.1.83
We suggest you to use the Anaconda/Miniconda to install all the packages.