Add Mac M1 Support

First thanks for this great job.

May I ask you to add “mps” support?

for the time being I did a “quick and dirty” fix adding

self.device= 'mps' if torch.backends.mps.is_available() == True else 'cpu'

after this line

https://github.com/asahi417/tner/blob/971bb6ff43c62bdfbb1f1793e69d15f4a592f90f/tner/ner_model.py#L83

and this

logging.info(f' GPUs is available? {torch.backends.mps.is_available()}')

after this

https://github.com/asahi417/tner/blob/971bb6ff43c62bdfbb1f1793e69d15f4a592f90f/tner/ner_model.py#L91

It doesn’t always work (for example with Roberta it doesn’t work but not because of T-Ner) but it seems to work most of the time.

Thanks

[asahi417]

I feel it’s safe to stick to the original one and put the M1 support as an optional, so in version 0.1.8, I reverted it to what used to be before.

Only if you need M1 support, please update the pytorch version accordingly so that your pytorch has the method torch.backends.mps and TNER will automatically switch to the M1 device if it’s available.

[asahi417]

@dpicca Hi, tner==0.1.7 is now supporting Mac M1. Please check it!