Creating tensors from numpy arrays with ">f" type yields bogus results

```
y = np.array([[1., 2., 3.],
             [4., 5., 6.]], dtype=">f")
torch.from_numpy(y)
```

Gives

```
1.00000e-41 *
  4.6006  0.0090  2.3049
  4.6007  5.7487  6.8966
[torch.FloatTensor of size 2x3]
```

Which is clearly incorrect. 

Such types are common when using `np.from_file`.