torch.cuda.is_available() is False
I am trying to load a pickle file of a pre-trained model to my code recipe but get the following error message:
I have already selected the "ai-exec-t4-gpu" in the "Containerized Execution" tab of the environment. I do not understand exactly what could be going wrong here.
Appreciate your help!
Best Answer
-
Sarina Dataiker, Dataiku DSS Core Designer, Dataiku DSS Adv Designer, Registered Posts: 317 DataikerHi @pbena64
,
Since there are several different configuration points that are relevant, it's indeed impossible to say without investigating a diagnostic. It does seem like the intended setup is that if you use one of these configurations you would have access to a GPU.
Thanks,
Sarina
Answers
-
Sarina Dataiker, Dataiku DSS Core Designer, Dataiku DSS Adv Designer, Registered Posts: 317 Dataiker
Hi @pbena64
,
There are a number of different configurations that are involved here. If you would like us to take a look, please open a support ticket with the following attached:
- an instance diagnostic of the DSS instance
- a job diagnostic of the job that fails with the referenced error
Then we can take a look.
Thank you,
Sarina -
Hi @SarinaS
,Thanks for the help. In my case, I don't see either of the "Actions > Download job diagnosis" and "Administration > Maintenance > Diagnostic tool." to download the logs (please see below screenshots). Could it be something with the configuration? Thanks!
Regards,
-
Sarina Dataiker, Dataiku DSS Core Designer, Dataiku DSS Adv Designer, Registered Posts: 317 Dataiker
Hi @pbena64
,
In that case, it appears that you are not an administrator of the DSS instance. Please reach out to your internal DSS administrator about the current issue. If they need any assistance troubleshooting, then they can open a support ticket with us the following attached:
- an instance diagnostic of the DSS instance
- a job diagnostic of the job that fails with the referenced error
Thanks!
Sarina -
Hi @SarinaS
,That is correct. I will reach out to the admins. Thanks a lot!
But, in general, if I select the "ai-exec-t4-gpu", or "ai-exec-t4-gpu-4g" from the "Containerized execution>>Build for" and save and update the environment, and in my recipe import torch and run torch.cuda.is_available(), I should get access to gpu, right? I mean no other setting has to be changed from my side?
Appreciate it!
-
Hi @SarinaS
,Thank you very much for your help. I will follow up with our internal administrator.
Regards,