search

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

Retraining/Fine tuning for different semantic parsing dataset #43

Closed pratikmjoshi closed 5 years ago

pratikmjoshi commented 5 years ago

Hi, Is there any way to currently use the pretrained models for inference for another semantic parsing task (NL->some other language)? If not, is it possible to fine tune the model somehow?

Thanks

bmccann commented 5 years ago

Yeah, it should be very doable as long as you can transform your semantic parsing task so that each example is of the form (context, question, answer). Then you can use the instructions from the Custom Inference section of the README.

For fine-tuning, you can actually do the same thing, but use the commands for training from the README instead of the commands for inference.