I’m curious about AutoNLP and have a question regarding what kind of models AutoNLP uses. e.g. I have a NER task, and from what I got on the documentation of the AutoNLP, - it will do search a best one. But let’s say, I have a domain-specific corpus with specific sets of labels, how it will be working in this case and what kind of model will be used? usual Bert model on NER hugginface pipeline? Is there way to customize a model and settings, e.g. vocabulary?
Hi with the new hub model training feature, you can do this now. Read more about it here: Training A Model From Hugging Face Hub — AutoNLP documentation
If you have any questions, please let me know.
Abhishek you’re awesome, sorry for the late response. This is a cool feature. Thank you! (I’m thinking about switching to AutoNLP for some projects and AutoNLP seems cool)
What base model to use for the NER task?
Hi @adrianog !
AutoNLP will choose the base model for you if you provide it the appropriate language
Under the “Training A Model From Hugging Face Hub” header I see:
$ autonlp create_project --name hub_model_training --task single_column_regression --hub_model abhishek/my_awesome_model --max_models 25
So I wondered whether I had to specify hub_model (abhishek/my_awesome_model).
But under Entity Extraction I see the project can be created with:
$ autonlp create_project --name entity_model --language en --task entity_extraction
so I assume that’s what you mean with “provide it the appropriate language” (–language param).
Can I pass a train file with NEW NER Labels? e.g. if I have a “chemical compound” NER to identify
Would it be possible for you to share a couple of your training examples? If not publicly, you can also share via email autonlp [at] huggingface [dot] co