oneflow.typing

Typing system

class oneflow.typing.Bundle

One or a collection of typing.Numpy/typing.ListNumpy/typing.ListListNumpy, such as x, [x], (x,), {“key”: x} and the mixed form of them.

class oneflow.typing.Callback

class oneflow.typing.ListListNumpy

ListListNumpy is a type hint for numpy output of a OneFlow global function For instance:

@oneflow.global_function()
def foo() -> oneflow.typing.ListListNumpy:
    mirrored_tensor_lists = ... # your network
    return mirrored_tensor_lists

mirrored_tensor_lists = foo() # get a list of list of numpy.ndarray
for tensor_list in mirrored_tensor_lists:
    for tensor in tensor_list:
        print(mirrored_tensors)

Placeholder(dtype=<class 'oneflow.python.framework.dtype.float32'>, batch_axis: Optional[int] = 0)

ListListNumpy.Placeholder is a typing function for numpy input of a OneFlow global function. A list of list of numpy.ndarray takes a ListListNumpy.Placeholder’s place. Each numpy.ndarray in the list could have any shape as long as it has the same rank and a smaller/equal size. For instance:

@oneflow.global_function()
def foo(
    image_blob: oneflow.typing.ListListNumpy.Placeholder(
        (2, 255, 255, 3), dtype=flow.float32
    )
):
    # your network

input1 = np.random.randn(2, 255, 255, 3).astype(np.float32)
input2 = np.random.randn(2, 251, 251, 3).astype(np.float32)
foo([[input1]])
foo([[input2]])

class oneflow.typing.ListNumpy

ListNumpy is a type hint for numpy output of a OneFlow global function For instance:

@oneflow.global_function()
def foo() -> oneflow.typing.ListNumpy:
    mirrored_tensors = ... # your network
    return mirrored_tensors

mirrored_tensors = foo() # get a list of numpy.ndarray
for tensor in mirrored_tensors:
    print(mirrored_tensors)

Placeholder(dtype=<class 'oneflow.python.framework.dtype.float32'>, batch_axis: Optional[int] = 0)

ListNumpy.Placeholder is a typing function for numpy input of a OneFlow global function. A list of numpy.ndarray takes a ListNumpy.Placeholder’s place. Each numpy.ndarray in the list could have any shape as long as it has the same rank and a smaller/equal size. For instance:

@oneflow.global_function()
def foo(
    image_blob: oneflow.typing.ListNumpy.Placeholder(
        (2, 255, 255, 3), dtype=flow.float32
    )
):
    # your network

input1 = np.random.randn(2, 255, 255, 3).astype(np.float32)
input2 = np.random.randn(2, 251, 251, 3).astype(np.float32)
foo([input1])
foo([input2])

class oneflow.typing.Numpy

Numpy is a type hint for numpy output of a OneFlow global function For instance:

@oneflow.global_function()
def foo() -> oneflow.typing.Numpy:
    loss = ... # your network
    return loss

loss = foo() # get a numpy.ndarray
print(loss)

Placeholder(dtype=<class 'oneflow.python.framework.dtype.float32'>, batch_axis: Optional[int] = 0)

Numpy.Placeholder is a typing function for numpy input of a OneFlow global function. A numpy.ndarray takes a Numpy.Placeholder’s place must have a identical shape. For instance:

@oneflow.global_function()
def foo(
    image_blob: oneflow.typing.Numpy.Placeholder(
        (2, 255, 255, 3), dtype=flow.float32
    )
):
    # your network

foo(np.random.randn(2, 255, 255, 3).astype(np.float32))