How to run the experiments of Orca when each server only has one GPU?

[SkyTu]

Hello, I am trying to run the experiments of Orca with two servers. Each server only possesses one GPU and I set the config as follows: (where private IP information was replaced as xxx.xxx.xxx.xxx)

{
    "P0": {
        "dealer": {
            "gpu": 0,
            "key_dir": "/tmp/"
        },
        "evaluator": {
            "gpu": 0,
            "peer": "127.0.0.1"
        }
    },
    "P1": {
        "dealer": {
            "gpu": 0,
            "key_dir": "/tmp/"
        },
        "evaluator": {
            "gpu": 0,
            "peer": "xxx.xxx.xxx.xxx(IP of P0)"
        }
    }
}

{
    "P0": {
        "dealer": {
            "gpu": 0,
            "key_dir": "/tmp/"
        },
        "evaluator": {
            "gpu": 0,
            "peer":  "xxx.xxx.xxx.xxx(IP of P1)"
        }
    },
    "P1": {
        "dealer": {
            "gpu": 0,
            "key_dir": "/tmp/"
        },
        "evaluator": {
            "gpu": 0,
            "peer": "127.0.0.1"
        }
    }
}

When I tried to execute "python3 run_experiment.py --table 3 --party 0" and "python3 run_experiment.py --table 3 --party 1", I met the fault "subprocess.CalledProcessError: Command 'CUDA_VISIBLE_DEVICES=0 ./orca_dealer 0 CNN2 /tmp/' returned non-zero exit status 127." How can I solve it?

[SkyTu]

I have solved the problem mentioned above. Since my Cuda version is 12.0 which is incompatible with the codes in ext package, I need to change it to version 11.7.