onnxruntime

mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-06-30 03:37:44 +00:00

History

Edward Chen e542cfd0e0 Introduce training changes.		2020-03-11 14:39:03 -07:00
..
docker	Introduce training changes.	2020-03-11 14:39:03 -07:00
scripts-ort	Introduce training changes.	2020-03-11 14:39:03 -07:00
scripts-pt	Introduce training changes.	2020-03-11 14:39:03 -07:00
nv_compare_job_entry.sh	Introduce training changes.	2020-03-11 14:39:03 -07:00
README.md	Introduce training changes.	2020-03-11 14:39:03 -07:00
real_train_job_entry.sh	Introduce training changes.	2020-03-11 14:39:03 -07:00
run.sh	Introduce training changes.	2020-03-11 14:39:03 -07:00
trend_job_entry.sh	Introduce training changes.	2020-03-11 14:39:03 -07:00
validation_job_entry.sh	Introduce training changes.	2020-03-11 14:39:03 -07:00

README.md

Compare ORT with NV PT Benchmark

Submit bash job with following job template

NV PT Job

{
    "version": "2019-10-30",
    "metadata": {
        "name": "pt_nv-baseline",
        "cluster": "rr3",
        "vc": "msrhyperscl"
    },
    "resources": {
        "workers": {
            "type": "skuResource",
            "sku": "G4",
            "count": 1,
            "image": "phillyregistry.azurecr.io/philly/jobs/custom/onnxruntime:v1",
            "commandLine": "$PHILLY_DATA_DIRECTORY/msrhyperscl/pengwa/profile/scripts-pt/profile-pt.sh \"fixed\"",
            "constraints": [
                {
                    "type": "uniqueConstraint",
                    "tag": "connectivityDomain"
                }
            ],
            "containerArgs": {
                "shmSize": "4G"
            }
        }
    }
}

Note:

G4 means choose the machine having 4 GPUs.

Comprehensive Batch Size Run

Submit bash job with following job template

NV PT Job

{
    "version": "2019-10-30",
    "metadata": {
        "name": "pt_nv-comprehensive_run",
        "cluster": "rr3",
        "vc": "msrhyperscl"
    },
    "resources": {
        "workers": {
            "type": "skuResource",
            "sku": "G4",
            "count": 1,
            "image": "phillyregistry.azurecr.io/philly/jobs/custom/onnxruntime:v1",
            "commandLine": "$PHILLY_DATA_DIRECTORY/msrhyperscl/pengwa/profile/scripts-pt/profile_trend-pt.sh",
            "constraints": [
                {
                    "type": "uniqueConstraint",
                    "tag": "connectivityDomain"
                }
            ],
            "containerArgs": {
                "shmSize": "4G"
            }
        }
    }
}

Note:

G4 means choose the machine having 4 GPUs.