ELMo calculates word embeddings based on sentences, so there are some difficulties when directly using the pre-trained embedding APIs from OpenNMT-py . Can you give me any advice on how to use ELMo under current OpenNMT ?
Yes, I am currently implementing this, along with a trainable language model that you can train to get ELMo embeddings in any language! I will do a Pull Request soon.
That's great 馃憤
Hi All,
any word on this project?
Hi All,
any word on this project?
Good question. The ELMo entry uses char in a CNN. OpenNMT-py accept bidirectional LSTM. I must review the changes on OpenNMT-py since 2018!
Most helpful comment
Yes, I am currently implementing this, along with a trainable language model that you can train to get ELMo embeddings in any language! I will do a Pull Request soon.