|
|
|
@@ -34,61 +34,56 @@ |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "NPU训练任务", |
|
|
|
"Name": "GPU调试任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"Cluster": "OpenI", |
|
|
|
"ComputeSource": "NPU", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "GPU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "ASCEND910", |
|
|
|
"CpuCores": 24, |
|
|
|
"MemGiB": 256, |
|
|
|
"AccCardType": "V100", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 50, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "-2", |
|
|
|
"ImageName": "mindspore2.0-cann6", |
|
|
|
"ImageUrl": "" |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "192.168.204.22:5000/default-workspace/99280a9940ae44ca8f5892134386fddb/image:ubuntu22.04-cuda11.8.0-py310-torch2.1.0-tf2.14.0" |
|
|
|
}, |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
"ID": "79942396-0f8d-4919-a490-a4b75a7da053", |
|
|
|
"ModelName": "Mindspore_MNIST_Example_Model" |
|
|
|
"ID": "f1711e48-65d6-436c-b34b-9698061979e7", |
|
|
|
"ModelName": "Torch_MNIST_Example_Model" |
|
|
|
} |
|
|
|
], |
|
|
|
"DatasetList": [ |
|
|
|
{ |
|
|
|
"ID": "aced78db-6703-41b6-be46-140ccd411107", |
|
|
|
"DatasetName": "MnistDataset_mindspore.zip" |
|
|
|
"ID": "8d5178b2-eb56-4c0d-b532-be8c685b1e61", |
|
|
|
"DatasetName": "MnistDataset_torch.zip" |
|
|
|
} |
|
|
|
], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "npu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "5" |
|
|
|
} |
|
|
|
] |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GPU训练任务", |
|
|
|
"Name": "GCU调试任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "GPU", |
|
|
|
"ComputeSource": "GCU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "V100", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 50, |
|
|
|
"AccCardType": "ENFLAME-I20", |
|
|
|
"CpuCores": 7, |
|
|
|
"MemGiB": 28, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "192.168.204.22:5000/default-workspace/99280a9940ae44ca8f5892134386fddb/image:ubuntu22.04-cuda11.8.0-py310-torch2.1.0-tf2.14.0" |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/enflame:i20_v1.2" |
|
|
|
}, |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
@@ -102,85 +97,75 @@ |
|
|
|
"DatasetName": "MnistDataset_torch.zip" |
|
|
|
} |
|
|
|
], |
|
|
|
"BranchName": "liuzx", |
|
|
|
"BootFile": "gpu_mnist_example/train.py", |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GPGPU训练任务", |
|
|
|
"Name": "MLU调试任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "ILUVATAR-GPGPU", |
|
|
|
"ComputeSource": "MLU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "BI-V100", |
|
|
|
"CpuCores": 30, |
|
|
|
"MemGiB": 64, |
|
|
|
"AccCardType": "MLU290", |
|
|
|
"CpuCores": 16, |
|
|
|
"MemGiB": 128, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/iluvatar:bi_corex3.1.1_pytorch1.13.1_py3.9_v1.0" |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/openioctopus/cambricon-pytorch:v1.0.7" |
|
|
|
}, |
|
|
|
"PretrainModelList": [], |
|
|
|
"DatasetList": [], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "gpgpu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "8" |
|
|
|
} |
|
|
|
] |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GCU训练任务", |
|
|
|
"Name": "DCU调试任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "GCU", |
|
|
|
"ComputeSource": "DCU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "ENFLAME-T20", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 30, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"AccCardType": "DCU", |
|
|
|
"CpuCores": 5, |
|
|
|
"MemGiB": 10, |
|
|
|
"GPUMemGiB": 10, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/enflame:t20_v1.0" |
|
|
|
"ImageID": "0b79fed7504c4e2dacpytorch21py387", |
|
|
|
"ImageName": "pytorch:2.1-centos7.6-dtk23.10-py38-ds0.12.3", |
|
|
|
"ImageUrl": "" |
|
|
|
}, |
|
|
|
"PretrainModelList": [], |
|
|
|
"DatasetList": [], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "gcu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "8" |
|
|
|
} |
|
|
|
] |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GPU调试任务", |
|
|
|
"Name": "GPGPU调试任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "GPU", |
|
|
|
"ComputeSource": "ILUVATAR-GPGPU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "V100", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 50, |
|
|
|
"AccCardType": "BI-V100", |
|
|
|
"CpuCores": 30, |
|
|
|
"MemGiB": 64, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "192.168.204.22:5000/default-workspace/99280a9940ae44ca8f5892134386fddb/image:ubuntu22.04-cuda11.8.0-py310-torch2.1.0-tf2.14.0" |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/iluvatar:bi_corex3.1.1_pytorch1.13.1_py3.9_v1.0" |
|
|
|
}, |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
@@ -199,119 +184,134 @@ |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GCU调试任务", |
|
|
|
"Name": "NPU训练任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "GCU", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"Cluster": "OpenI", |
|
|
|
"ComputeSource": "NPU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "ENFLAME-I20", |
|
|
|
"CpuCores": 7, |
|
|
|
"MemGiB": 28, |
|
|
|
"AccCardType": "ASCEND910", |
|
|
|
"CpuCores": 24, |
|
|
|
"MemGiB": 256, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/enflame:i20_v1.2" |
|
|
|
"ImageID": "-2", |
|
|
|
"ImageName": "mindspore2.0-cann6", |
|
|
|
"ImageUrl": "" |
|
|
|
}, |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
"ID": "f1711e48-65d6-436c-b34b-9698061979e7", |
|
|
|
"ModelName": "Torch_MNIST_Example_Model" |
|
|
|
"ID": "79942396-0f8d-4919-a490-a4b75a7da053", |
|
|
|
"ModelName": "Mindspore_MNIST_Example_Model" |
|
|
|
} |
|
|
|
], |
|
|
|
"DatasetList": [ |
|
|
|
{ |
|
|
|
"ID": "8d5178b2-eb56-4c0d-b532-be8c685b1e61", |
|
|
|
"DatasetName": "MnistDataset_torch.zip" |
|
|
|
"ID": "aced78db-6703-41b6-be46-140ccd411107", |
|
|
|
"DatasetName": "MnistDataset_mindspore.zip" |
|
|
|
} |
|
|
|
], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
"BootFile": "npu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "5" |
|
|
|
} |
|
|
|
] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "MLU调试任务", |
|
|
|
"Name": "GPU训练任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "MLU", |
|
|
|
"ComputeSource": "GPU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "MLU290", |
|
|
|
"CpuCores": 16, |
|
|
|
"MemGiB": 128, |
|
|
|
"AccCardType": "V100", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 50, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/openioctopus/cambricon-pytorch:v1.0.7" |
|
|
|
"ImageUrl": "192.168.204.22:5000/default-workspace/99280a9940ae44ca8f5892134386fddb/image:ubuntu22.04-cuda11.8.0-py310-torch2.1.0-tf2.14.0" |
|
|
|
}, |
|
|
|
"PretrainModelList": [], |
|
|
|
"DatasetList": [], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "", |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
"ID": "f1711e48-65d6-436c-b34b-9698061979e7", |
|
|
|
"ModelName": "Torch_MNIST_Example_Model" |
|
|
|
} |
|
|
|
], |
|
|
|
"DatasetList": [ |
|
|
|
{ |
|
|
|
"ID": "8d5178b2-eb56-4c0d-b532-be8c685b1e61", |
|
|
|
"DatasetName": "MnistDataset_torch.zip" |
|
|
|
} |
|
|
|
], |
|
|
|
"BranchName": "liuzx", |
|
|
|
"BootFile": "gpu_mnist_example/train.py", |
|
|
|
"Parameters": [] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "DCU调试任务", |
|
|
|
"Name": "GPGPU训练任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "DCU", |
|
|
|
"ComputeSource": "ILUVATAR-GPGPU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "DCU", |
|
|
|
"CpuCores": 5, |
|
|
|
"MemGiB": 10, |
|
|
|
"GPUMemGiB": 10, |
|
|
|
"AccCardType": "BI-V100", |
|
|
|
"CpuCores": 30, |
|
|
|
"MemGiB": 64, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "0b79fed7504c4e2dacpytorch21py387", |
|
|
|
"ImageName": "pytorch:2.1-centos7.6-dtk23.10-py38-ds0.12.3", |
|
|
|
"ImageUrl": "" |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/iluvatar:bi_corex3.1.1_pytorch1.13.1_py3.9_v1.0" |
|
|
|
}, |
|
|
|
"PretrainModelList": [], |
|
|
|
"DatasetList": [], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
"BootFile": "gpgpu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "8" |
|
|
|
} |
|
|
|
] |
|
|
|
}, |
|
|
|
{ |
|
|
|
"Name": "GPGPU调试任务", |
|
|
|
"Name": "GCU训练任务", |
|
|
|
"Description": "", |
|
|
|
"JobType": "DEBUG", |
|
|
|
"JobType": "TRAIN", |
|
|
|
"Cluster": "C2Net", |
|
|
|
"ComputeSource": "ILUVATAR-GPGPU", |
|
|
|
"ComputeSource": "GCU", |
|
|
|
"HasInternet": 2, |
|
|
|
"AccCardsNum": 1, |
|
|
|
"AccCardType": "BI-V100", |
|
|
|
"CpuCores": 30, |
|
|
|
"MemGiB": 64, |
|
|
|
"AccCardType": "ENFLAME-T20", |
|
|
|
"CpuCores": 8, |
|
|
|
"MemGiB": 30, |
|
|
|
"GPUMemGiB": 32, |
|
|
|
"ShareMemGiB": 0, |
|
|
|
"Image": { |
|
|
|
"ImageID": "", |
|
|
|
"ImageName": "", |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/iluvatar:bi_corex3.1.1_pytorch1.13.1_py3.9_v1.0" |
|
|
|
"ImageUrl": "swr.cn-south-1.myhuaweicloud.com/grampus/enflame:t20_v1.0" |
|
|
|
}, |
|
|
|
"PretrainModelList": [ |
|
|
|
{ |
|
|
|
"ID": "f1711e48-65d6-436c-b34b-9698061979e7", |
|
|
|
"ModelName": "Torch_MNIST_Example_Model" |
|
|
|
} |
|
|
|
], |
|
|
|
"DatasetList": [ |
|
|
|
"PretrainModelList": [], |
|
|
|
"DatasetList": [], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "gcu_mnist_example/train.py", |
|
|
|
"Parameters": [ |
|
|
|
{ |
|
|
|
"ID": "8d5178b2-eb56-4c0d-b532-be8c685b1e61", |
|
|
|
"DatasetName": "MnistDataset_torch.zip" |
|
|
|
"Label": "epoch_size", |
|
|
|
"Value": "8" |
|
|
|
} |
|
|
|
], |
|
|
|
"BranchName": "master", |
|
|
|
"BootFile": "", |
|
|
|
"Parameters": [] |
|
|
|
] |
|
|
|
} |
|
|
|
] |