How to use t4 GPU from GCP

[mehadi92]

I'm using this command for t4 GPU in yml

cml runner \
              --cloud=gcp \
              --cloud-region=us-central1-a \
              --cloud-type=n1-highmem-2 \
              --labels=cml-gpu \
              --single=true \
              --cloud-spot=false \
              --cloud-gpu=t4 \
              --cloud-hdd-size=50

But getting this issue

  Argument: cloud-gpu, Given: "t4", Choices: "nogpu", "k80", "v100", "tesla"

I also try with "tesla" and it works but it takes tesla v100.

How can I use tesla t4 GPU

Thanks

[0x2b3bfa0]

Good question, @mehadi92. 😅 After releasing #1049 you should be able to use --cloud-gpu=nvidia-tesla-t4 or any other supported value.

[0x2b3bfa0]

In the meanwhile, you can test it with npx github:iterative/cml#native-gpu runner --cloud=gcp ···

[mehadi92]

@0x2b3bfa0 It would be better If you could provide a list of supportd values. Otherwise, it's easy to fall down into naming confusion - Thanks

[0x2b3bfa0]

It supports any accelerator available on Google Cloud for the chosen instance type. See also gcloud compute accelerator-types and the https://cloud.google.com/compute/docs/gpus documentation.