Flair: Is it possible to disable CUDA in some or all of the library?

[prematurelyoptimized]

I am trying to run the Flair framework through its paces on my machine, but my CUDA stack is currently not working. Thus, when I try even the example usage in the README,

from flair.data import Sentence
from flair.models import SequenceTagger

# make a sentence
sentence = Sentence('I love Berlin .')

# load the NER tagger
tagger = SequenceTagger.load('ner')

# run NER over sentence
tagger.predict(sentence)

CUDA has a heart attack and dies.

Is there a way that I can set a flag or configuration value to force the Flair methods to use the CPU? I know that some modules won't work on the CPU (huggingface's implementation of BERT among them), but I would think that most functionality would work without GPU support, just slowly.

Patience I have. A functioning GPU...... not yet.

[stefan-it]

Hi @prematurelyoptimized ,

try to set the following environment variable

export CUDA_VISIBLE_DEVICES=""

before calling your Python script. That should work :)

[alanakbik]

Hello @prematurelyoptimized another option is to set the flair.device at the beginning of your python script, like this:

import torch, flair

# first, set your device
flair.device =  torch.device("cpu")

# then do your normal code
from flair.data import Sentence
from flair.embeddings import FlairEmbeddings

FlairEmbeddings('news-forward').embed(Sentence('Hello World'))

...

[prematurelyoptimized]

Since I'm running in a Jupyter notebook, Alan's approach is simpler to implement and works perfectly. Good to know about the environment variable though. Thank you much.