Function Differences with torch.Tensor.float

torch.Tensor.float

torch.Tensor.float(memory_format=torch.preserve_format)

For more information, see torch.Tensor.float.

mindspore.ops.Cast

class mindspore.ops.Cast(*args, **kwargs)(
    input_x,
    type
)

For more information, see mindspore.ops.Cast.

Differences

PyTorch: Changes the tensor type to float.

MindSpore：Converts the input type to the specified data type.

Code Example

import mindspore as ms
import mindspore.ops as ops
import torch
import numpy as np

# In MindSpore, you can specify the data type to be transformed into.
input_x = ms.Tensor(np.random.randn(2, 3, 4, 5).astype(np.float32))
cast = ops.Cast()
output = cast(input_x, ms.int32)
print(output.dtype)
print(output.shape)
# Out：
# Int32
# (2, 3, 4, 5)

# In torch, the input will be transformed into float.
input_x = torch.Tensor(np.random.randn(2, 3, 4, 5).astype(np.int32))
output = input_x.float()
print(output.dtype)
print(output.shape)
# Out：
# torch.float32
# torch.Size([2, 3, 4, 5])