Fix padding for int contexts (#227)
*Issue #, if available:* On Linux, the final call to `.to` creates
trouble when input tensors are integer. For example:
```
>>> a = torch.tensor([1])
>>> b = torch.stack([torch.full((1,), torch.nan), a])
>>> b
tensor([[nan],
[1.]])
>>> b.to(a)
tensor([[-9223372036854775808],
[ 1]])
```
By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.
2024-12-04 15:46:17 +00:00
|
|
|
# SPDX-License-Identifier: Apache-2.0
|
