microsoft/hi-ml

Bug: Jobs submitted to a Kubernetes compute target don't use the GPUs

ant0nsc opened this issue · 0 comments

Jobs that work have this section in the run definition:

        "cmk8sCompute": {
            "configuration": {
                "gpu_count": 16,
                "preemption_allowed": false,
                "job_priority": 200,
                "enable_ssh": true,
                "ssh_public_key": "
            }
        },