salesforce / decaNLP

The Natural Language Decathlon: A Multitask Challenge for NLP
BSD 3-Clause "New" or "Revised" License
2.34k stars 474 forks source link

Training on Custom Dataset #47

Closed ashleyyy94 closed 5 years ago

ashleyyy94 commented 5 years ago

I have a dataset in the format of context, question, and answer. Besides being able to run Inference on a Custom Dataset, is it possible to train a new model on our own dataset as well?

Thank you.

keskarnitish commented 5 years ago

This should work. If you add the dataset in the data folder, say, .data/mydata/ with two files: train.jsonl and val.jsonl, you should be able to pass in --train_tasks mydata for training on that.

Closing for now, feel free to reopen if your issue wasn't resolved.