Closed zerogerc closed 3 years ago
@zerogerc you can achieve this by passing 'requires_grad=True' when initializing the BertEmbedding. However, the code for a model without LSTM layers/word embeddings is not integrated into the repo yet. I'm working on this, and it can be available in the next release.
Thanks!
Hi, thanks for the awesome library. I wonder if it is possible to pretrain the model on one dataset and then fine-tune on another?