randint accept generator=None (#29748)
Summary:
This PR fixes the inconsistent behavior of `randint`'s `generator=` kwarg. It does not accept `None`, which is inconsistent with how other random functions behave:
```
In [12]: torch.randint(0, 4, size=(2,3), generator=torch.Generator())
Out[12]:
tensor([[2, 0, 1],
[0, 1, 3]])
In [13]: torch.randint(0, 4, size=(2,3), generator=None)
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-13-a6bc6525a1e1> in <module>
----> 1 torch.randint(0, 4, size=(2,3), generator=None)
TypeError: randint() received an invalid combination of arguments - got (int, int, generator=NoneType, size=tuple), but expected one of:
* (int high, tuple of ints size, torch.Generator generator, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool requires_grad)
* (int high, tuple of ints size, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool requires_grad)
* (int low, int high, tuple of ints size, torch.Generator generator, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool requires_grad)
* (int low, int high, tuple of ints size, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool requires_grad)
```
Other random functions work fine:
```
In [9]: torch.bernoulli(torch.ones(3))
Out[9]: tensor([1., 1., 1.])
In [10]: torch.bernoulli(torch.ones(3), generator=None)
Out[10]: tensor([1., 1., 1.])
```
This PR also documents the `generator=` kwarg, and fixes https://github.com/pytorch/pytorch/issues/29683 since it's a related easy fix.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29748
Differential Revision: D18529951
Pulled By: ezyang
fbshipit-source-id: e956cc989decc94e9483fd4a30f9255240d7c07e
