Allennlp: GPU for Named Entity Recognition

[saurabhbh21]

I'm trying to execute the NER model (https://s3-us-west-2.amazonaws.com/allennlp/models/ner-model-2018.04.26.tar.gz) in GPU, but it's not using GPU(rather normal CPU) in Azure NV6.

[joelgrus]

what's the exact command you're running?

[VictorSanh]

I sometimes have to add CUDA_VISIBLE_DEVICES=0 at the beginning of the bash command (and replace 0 with whichever GPU you want to use). I don't know if it can help....

[saurabhbh21]

from allennlp.predictors import Predictor

model_url = "https://s3-us-west-2.amazonaws.com/allennlp/models/ner-model-2018.04.26.tar.gz"
predictor = Predictor.from_path(model_path)
result = predictor.predict(sentence=some_text)

What changes should I do (in python)?

[nelson-liu]

as a fairly hacky solution, you can use:

model_url = "s3-us-west-2.amazonaws.com/allennlp/models/ner-model-2018.04.26.tar.gz"
predictor = Predictor.from_path(model_path)
predictor._model = predictor._model.cuda()
result = predictor.predict(sentence=some_text)

[joelgrus]

Predictor.from_path should really take a cuda_device parameter, I'll open an issue for that.

In the meantime, just do

archive = load_archive(model_path, cuda_device=0)
predictor = Predictor.from_archive(archive)

(or use Nelson's hack)

[saurabhbh21]

Thanx for the help.