Fix eager optim dynamic attr bug

[wyg1997]

修复 https://github.com/Oneflow-Inc/OneTeam/issues/564#issuecomment-908333150 提到的问题

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 128.5ms (= 6423.2ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 142.0ms (= 7098.7ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.11 (= 142.0ms / 128.5ms)

OneFlow resnet50 time: 74.6ms (= 3731.1ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 83.2ms (= 4160.4ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.12 (= 83.2ms / 74.6ms)

OneFlow resnet50 time: 49.6ms (= 2480.7ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 57.4ms (= 2869.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.16 (= 57.4ms / 49.6ms)

OneFlow resnet50 time: 46.0ms (= 2300.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 47.2ms (= 2360.7ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.03 (= 47.2ms / 46.0ms)

OneFlow resnet50 time: 39.6ms (= 1977.9ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 44.9ms (= 2246.5ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.14 (= 44.9ms / 39.6ms)

OneFlow resnet50 time: 140.7ms (= 7033.3ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 149.6ms (= 7482.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.06 (= 149.6ms / 140.7ms)

OneFlow resnet50 time: 87.5ms (= 4374.0ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 91.5ms (= 4574.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.05 (= 91.5ms / 87.5ms)

OneFlow resnet50 time: 58.6ms (= 2931.2ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 61.7ms (= 3083.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 1.05 (= 61.7ms / 58.6ms)

OneFlow resnet50 time: 56.4ms (= 2819.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 53.4ms (= 2668.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 0.95 (= 53.4ms / 56.4ms)

OneFlow resnet50 time: 60.1ms (= 3004.4ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow GPU used (rank 0): 0 MiB
PyTorch resnet50 time: 47.6ms (= 2381.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
PyTorch GPU used (rank 0, estimated): 0 MiB
Relative speed: 0.79 (= 47.6ms / 60.1ms)