Open Sreyan88 opened 1 year ago
Hi, thanks for your interest! For the NER task, we directly use the pre-trained genius-large
model, without fine-tuning. If you want to fine-tune the model for NER, you can modify the finetuning script for classification task. I didn't remember the concrete training time (maybe several minutes for hendreds of samples?)
I try to fine-tune with Genius-f on a custom NER dataset and it seems to converge in about 2 epochs (lowest eval loss). Is this behavior normal?
Hi there! Great repo. Would you mind sharing the logs for Genius-f fine-tuning for the NER use case? Or would you mind giving an estimate of the final rouge metrics and how much time it took to converge?
Thank You!