[numpy] torch.exp{2, m1}: promote integer inputs to float

[kshitij12345]

Reference: #42515

[kshitij12345]

Reason for skipping for bfloat16

>>> import torch
>>> torch.tensor(6.75, dtype=torch.bfloat16)
tensor(6.7500, dtype=torch.bfloat16)
>>> t = torch.tensor(6.75, dtype=torch.bfloat16)
>>> torch.expm1(t)
tensor(852., dtype=torch.bfloat16)
>>> torch.expm1(t.to(torch.float32))
tensor(853.0588)

There are cases with more difference in tolerance for other values.

[dr-ci]

💊 CI failures summary and remediations

As of commit ced829a (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

1 failure not recognized by patterns:

Job	Step	Action
pytorch_linux_xenial_py3_clang5_asan_test2	Report results	🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 1 time.

[mruberry]

Reason for skipping for bfloat16

>>> import torch
>>> torch.tensor(6.75, dtype=torch.bfloat16)
tensor(6.7500, dtype=torch.bfloat16)
>>> t = torch.tensor(6.75, dtype=torch.bfloat16)
>>> torch.expm1(t)
tensor(852., dtype=torch.bfloat16)
>>> torch.expm1(t.to(torch.float32))
tensor(853.0588)

There are cases with more difference in tolerance for other values.

This makes sense. I would really like to start considering ways we can work around NumPy's lack of bfloat16 support. Maybe we can cast the values to float32, then cast them to bfloat16, then nextafter them in the direction of the PyTorch result?

[mruberry]

Awesome! Thanks @kshitij12345!

[facebook-github-bot]

@mruberry has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[kshitij12345]

This makes sense. I would really like to start considering ways we can work around NumPy's lack of bfloat16 support. Maybe we can cast the values to float32, then cast them to bfloat16, then nextafter them in the direction of the PyTorch result?

I think that makes sense since both have same range. Right now the catch is that nextafter is not implemented for bfloat16.

[facebook-github-bot]

@mruberry merged this pull request in eb9516e.