I have a text_classifier_learner which I use to classify text based on an ULMFit non-English language model I have pretrained and fine-tuned myself. I noticed that when I create the text_classifier_learner to train it (classifier = text_classifier_learner(my_data, AWD_LSTM, drop_mult=0.5)), fastai downloads https://s3.amazonaws.com/fast-ai-modelzoo/wt103-fwd.tgz and I’m wondering why this happens…
As far as I understand, this is the pretrained ULMFit wikipedia-103 English language model, correct?
If that is the case, why would I use this instead of my own pretrained language model, which should have the same structure?