Decouple user_kernel and device_tag

[liujuncheng]

No description provided.

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 129.2ms (= 12922.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.6ms (= 14364.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 143.6ms / 129.2ms)

OneFlow resnet50 time: 75.8ms (= 7581.0ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 86.6ms (= 8658.6ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.14 (= 86.6ms / 75.8ms)

OneFlow resnet50 time: 48.6ms (= 9712.0ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.8ms (= 12166.2ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.25 (= 60.8ms / 48.6ms)

OneFlow resnet50 time: 41.6ms (= 8317.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 41.3ms (= 8250.9ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 0.99 (= 41.3ms / 41.6ms)

OneFlow resnet50 time: 34.7ms (= 6936.1ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 35.4ms (= 7087.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.02 (= 35.4ms / 34.7ms)

OneFlow swin dataloader time: 0.276s (= 55.260s / 200, num_workers=1)
PyTorch swin dataloader time: 0.146s (= 29.224s / 200, num_workers=1)
Relative speed: 0.529 (= 0.146s / 0.276s)

OneFlow swin dataloader time: 0.078s (= 15.614s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.479s / 200, num_workers=4)
Relative speed: 0.543 (= 0.042s / 0.078s)

OneFlow swin dataloader time: 0.043s (= 8.605s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.370s / 200, num_workers=8)
Relative speed: 0.508 (= 0.022s / 0.043s)

❌ OneFlow resnet50 time: 145.2ms (= 14522.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 169.8ms (= 16981.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 169.8ms / 145.2ms)

OneFlow resnet50 time: 93.9ms (= 9394.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 113.4ms (= 11338.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.21 (= 113.4ms / 93.9ms)

OneFlow resnet50 time: 67.9ms (= 13586.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 98.0ms (= 19590.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.44 (= 98.0ms / 67.9ms)

OneFlow resnet50 time: 56.7ms (= 11345.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.5ms (= 15091.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 75.5ms / 56.7ms)

OneFlow resnet50 time: 50.3ms (= 10053.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.2ms (= 13847.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 69.2ms / 50.3ms)

[github-actions]

CI failed when running job: cuda-module. PR label automerge has been removed

[github-actions]

Speed stats:

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8529/

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 129.2ms (= 12918.9ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 150.1ms (= 15005.2ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.16 (= 150.1ms / 129.2ms)

OneFlow resnet50 time: 75.9ms (= 7593.6ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.8ms (= 8575.6ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.13 (= 85.8ms / 75.9ms)

OneFlow resnet50 time: 50.0ms (= 9990.7ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.3ms (= 12063.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.21 (= 60.3ms / 50.0ms)

OneFlow resnet50 time: 40.8ms (= 8166.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 45.0ms (= 8995.1ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.10 (= 45.0ms / 40.8ms)

OneFlow resnet50 time: 36.1ms (= 7213.6ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.3ms (= 7662.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.06 (= 38.3ms / 36.1ms)

OneFlow swin dataloader time: 0.278s (= 55.680s / 200, num_workers=1)
PyTorch swin dataloader time: 0.150s (= 29.901s / 200, num_workers=1)
Relative speed: 0.537 (= 0.150s / 0.278s)

OneFlow swin dataloader time: 0.074s (= 14.778s / 200, num_workers=4)
PyTorch swin dataloader time: 0.040s (= 7.913s / 200, num_workers=4)
Relative speed: 0.535 (= 0.040s / 0.074s)

OneFlow swin dataloader time: 0.041s (= 8.153s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.544s / 200, num_workers=8)
Relative speed: 0.557 (= 0.023s / 0.041s)

❌ OneFlow resnet50 time: 144.4ms (= 14438.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 170.3ms (= 17032.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.18 (= 170.3ms / 144.4ms)

OneFlow resnet50 time: 93.6ms (= 9363.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 116.9ms (= 11692.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 116.9ms / 93.6ms)

OneFlow resnet50 time: 70.0ms (= 14005.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 87.9ms (= 17573.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 87.9ms / 70.0ms)

OneFlow resnet50 time: 56.8ms (= 11352.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.7ms (= 15333.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.35 (= 76.7ms / 56.8ms)

OneFlow resnet50 time: 47.9ms (= 9584.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.0ms (= 15199.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.59 (= 76.0ms / 47.9ms)