-
-
Notifications
You must be signed in to change notification settings - Fork 5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Array API failures with Metal Performance Shaders/torch #20700
Comments
On the other hand, https://data-apis.org/array-api/latest/API_specification/data_types.html is pretty clear on requiring support for those types, so we'd likely only want to shim around cases where a violation of the standard doesn't occur I suppose. Test skips could be added with clear explanations if we thought this backend merited it. |
I think that guidance in the array API standard may be loosened a bit. However, in SciPy we do require |
With latest
main
branch on an ARM macSCIPY_DEVICE=mps python dev.py -n test -j 16 -b all
results in 189 failures.Some of the failures will be hard/impossible to avoid because
torch
doesn't support all ops on that device just yet. However, many of the failures do look avoidable givenTypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.
is probably something that could be shimmed around in some places perhaps.Short test summary info only for now
TypeError: Trying to convert ComplexDouble to the MPS backend but it does not have support for that dtype.
might be hard to avoid too.Using
torch
2.3.0
.The text was updated successfully, but these errors were encountered: