Make tests succeed more on MPS #1463
Conversation
|
cc @BenjaminBossan :) |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
There was a problem hiding this comment.
Thanks a lot @akx ! We should be able to merge after fixing the styling checks ! 🙏 Can you format your changes ?
|
@younesbelkada Done... Can you maybe take a peek at #1467 so I don't need to manually remember to run |
|
|
||
|
|
||
| class PeftAutoModelTester(unittest.TestCase): | ||
| dtype = torch.float16 if infer_device() == "mps" else torch.bfloat16 |
There was a problem hiding this comment.
Would it be possible to check if torch.cuda.is_bf16_supported() instead or does that not work on Mac?
There was a problem hiding this comment.
I wouldn't trust anything that says cuda on the tin to work with Macs 😁
There was a problem hiding this comment.
Maybe that could be held off for later, anyway?
|
Thanks @akx for helping out all the Mac users out there :) |
Some tests would not succeed on MPS devices because
bfloat16is not supported. This makes them get skipped instead.On a similar note, tests that are not applicable (e.g. generated via a
parametrizedmatrix and should have been skipped) no longer seem to quietly pass, but are instead markedskip.It also fixes up the rtype for
infer_deviceto always be a string – previously, it would return a string, or in the case ofmps, atorch.device().Sibling of #1448, since I got frustrated trying to run tests on my Apple Silicon device and figuring which ones of them are actual failures.