📚 The doc issue

The doc explains d parameter but it only says Either torch.float32 or torch.float64 as shown below:

d (torch.dtype) – the floating point dtype to make the default. Either torch.float32 or torch.float64.

Actually, d parameter with torch.float16, torch.half, torch.bfloat16, torch.float or torch.double type also works as shown below:

import torch

torch.set_default_dtype(d=torch.float16)
torch.set_default_dtype(d=torch.half)
torch.get_default_dtype()
# torch.float16

torch.set_default_dtype(d=torch.bfloat16)
torch.get_default_dtype()
# torch.bfloat16

torch.set_default_dtype(d=torch.float32)
torch.set_default_dtype(d=torch.float)
torch.get_default_dtype()
# torch.float32

torch.set_default_dtype(d=torch.float64)
torch.set_default_dtype(d=torch.double)
torch.get_default_dtype()
# torch.float64

Suggest a potential alternative/fix

So, Either torch.float32 or torch.float64 should be removed from the doc should say int type as shown below:

d (torch.dtype) – the floating point dtype to make the default.

cc @svekars @brycebortree @albanD

Sure, we'd accept a PR for tweaking the docs to match what's actually supported.

@hyperkai
The doc link leads to set_default_device instead of set_default_dtype