LazyInterpret build LocalTensor if input is local

[chengtbf]

LazyInterpret 在推导 output Tensor 时，根据 input tensor 是否是 Local 来创建 LocalTensor 或者 ConsistentTensor。
但是 LazyInterpret 调用 JobBuildAndInferCtx AddOp 时，永远是 ConsistentOp。

[github-actions]

CI failed, removing label automerge

[github-actions]

CI failed, removing label automerge

[github-actions]

CI failed, removing label automerge

[strint]

ConsistentOp支持输入Lazy Local Tensor？ 看这里把loca 的input和variable都转成了Local，而不是单卡的consistent

[strint]

ConsistentOp支持输入Lazy Local Tensor？ 看这里把loca 的input和variable都转成了Local，而不是单卡的consistent

[chengtbf]

ConsistentOp支持输入Lazy Local Tensor？ 看这里把loca 的input和variable都转成了Local，而不是单卡的consistent

ConsistentOp 跟 Tensor 没有任何关系。ConsistentOp 不需要 Tensor 这个概念，只需要 lbn 就行。

[strint]

ConsistentOp支持输入Lazy Local Tensor？ 看这里把loca 的input和variable都转成了Local，而不是单卡的consistent

ConsistentOp 跟 Tensor 没有任何关系。ConsistentOp 不需要 Tensor 这个概念，只需要 lbn 就行。

哦，这些Tensor只是存了一些必要的meta信息，方便在job外以一个统一的数据结构交换meta信息的。

[chengtbf]

ConsistentOp支持输入Lazy Local Tensor？ 看这里把loca 的input和variable都转成了Local，而不是单卡的consistent

ConsistentOp 跟 Tensor 没有任何关系。ConsistentOp 不需要 Tensor 这个概念，只需要 lbn 就行。

哦，这些Tensor只是存了一些必要的meta信息，方便在job外以一个统一的数据结构交换meta信息的。

是的~

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.1ms (= 6956.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 126.1ms (= 6305.6ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 139.1ms / 126.1ms)

PyTorch resnet50 time: 84.2ms (= 4210.9ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 73.6ms (= 3682.2ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 84.2ms / 73.6ms)

PyTorch resnet50 time: 58.2ms (= 2912.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.6ms (= 2528.5ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.15 (= 58.2ms / 50.6ms)

PyTorch resnet50 time: 49.7ms (= 2486.6ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.2ms (= 2359.0ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.05 (= 49.7ms / 47.2ms)

PyTorch resnet50 time: 44.1ms (= 2206.9ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 46.6ms (= 2332.0ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.95 (= 44.1ms / 46.6ms)

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.9ms (= 6994.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 125.9ms (= 6293.2ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 139.9ms / 125.9ms)

PyTorch resnet50 time: 83.5ms (= 4172.5ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 73.7ms (= 3684.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 83.5ms / 73.7ms)

PyTorch resnet50 time: 58.3ms (= 2915.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 49.6ms (= 2480.3ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.18 (= 58.3ms / 49.6ms)

PyTorch resnet50 time: 48.9ms (= 2446.0ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 46.8ms (= 2337.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.05 (= 48.9ms / 46.8ms)

PyTorch resnet50 time: 44.8ms (= 2239.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.0ms (= 2348.0ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.95 (= 44.8ms / 47.0ms)