search

allenai / allennlp

An open-source NLP research library, built on PyTorch.
http://www.allennlp.org
Apache License 2.0
11.75k stars 2.25k forks source link

can we fine tune SRL model using our own data? #2865

Closed omerarshad closed 5 years ago

omerarshad commented 5 years ago

is there a way to finetune pretrained allennlp SRL model using our own data?

joelgrus commented 5 years ago

yes, there is a fine-tune command:

https://github.com/allenai/allennlp/blob/master/allennlp/commands/fine_tune.py

that just needs the path to the trained model archive, a config file specifying the details of the fine tuning (what data to use, what training parameters to use, etc) and a directory in which to put the outputs

omerarshad commented 5 years ago

and what should be the format of data? should it be conll2012 format?

matt-gardner commented 5 years ago

This one is actually tricky, because the SRL model depends on old components. @DeNeutoy opened a PR with a BERT-based model, which would be better to use today.

omerarshad commented 5 years ago

so any link to it? and is it implemented? secondly what if i have data in BIO tag format?

omerarshad commented 5 years ago

any progress?

shanalikhan commented 4 years ago

@joelgrus

that just needs the path to the trained model archive, a config file specifying the details of the fine tuning (what data to use, what training parameters to use, etc) and a directory in which to put the outputs

Can you provide some example of fine-tune config regarding bert or any other model, it would be very helpful?