Remove hard-coded dtype from best_f buffers

[saitcakmak]

Summary:
Specifiying dtype=float (equivalent to torch.float64) causes issues if the user wants to use single precision. See #2724

Removing the dtype argument from torch.as_tensor will lead to using the existing dtype if the input is a tensor, and using torch.get_default_dtype() if the input is a python float or a list of floats. If the input is a numpy array, the corresponding torch dtype is used (e.g. torch.float64 for np.float64).

Differential Revision: D69121327

[Balandat]

One thing to note here is that if the input is a python list with ints then this will register an int tensor which will cause breakages downstream. This shouldn't happen a lot but I think we should auto-convert the dtype in that case.

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D69121327

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.99%. Comparing base (3faf489) to head (fca8bd5).
Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2725   +/-   ##
=======================================
  Coverage   99.99%   99.99%           
=======================================
  Files         203      203           
  Lines       18671    18671           
=======================================
  Hits        18670    18670           
  Misses          1        1           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[saitcakmak]

One thing to note here is that if the input is a python list with ints then this will register an int tensor which will cause breakages downstream. This shouldn't happen a lot but I think we should auto-convert the dtype in that case.

Yeah, that's a fair point. I can't find a clean way to convert to the default float type without explicitly checking for integers and casting after, which I don't like. The type hints specify float | Tensor. I think it's ok to expect the user to respect that and specify an appropriate inputs. I'm sure various other parts of the code will also break if you pass in integer tensors.

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D69121327

[facebook-github-bot]

This pull request has been merged in c0db823.