Lesson 2 - troubleshoot macbook m1 issue

Part 1 2022
1

Has anyone managed to solve the error (see below) that arises when attempting to use aug_transforms technique on macos m1?

NotImplementedError: The operator 'aten::_linalg_solve_ex.result' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature, please comment on https://github.com/pytorch/pytorch/issues/77764. As a temporary fix, you can set the environment variable `PYTORCH_ENABLE_MPS_FALLBACK=1` to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.

Update: using chatgpt4 I learned that adding below adjustment helps a little bit:

[instanceOfDataBlock].new(item_tfms=Resize(128), batch_tfms=custom_aug_transforms)

where custom_aug_transforms is

from fastai.vision.augment import *
custom_aug_transforms = [RandomResizedCrop(128, min_scale=0.35), Flip(), Brightness(), Contrast(), Rotate(max_deg=10.0)]

I am not sure impact of custom_aug_transforms is equally good as aug_transforms but seems not too bad. In the meantime please visit github issue to implement aten::_linalg_solve_ex.result operator and vote for it to increase its priority

2 Likes
3

Here is how I got the code to work:

from fastai.vision.augment import *
custom_aug_transforms = [RandomResizedCrop(128, min_scale=0.35), Flip(), Brightness(), Contrast(), Rotate(max_deg=10.0)]

bears = bears.new(item_tfms=Resize(128), batch_tfms=custom_aug_transforms)
dls = bears.dataloaders(path)
dls.train.show_batch(max_n=8, nrows=2, unique=True)

4

Don’t forget to change the following cell to:
bears = bears.new(
item_tfms=RandomResizedCrop(224, min_scale=0.5),
batch_tfms=custom_aug_transforms)
dls = bears.dataloaders(path)

5

I tried this and I still get
TypeError: can’t convert mps:0 device type tensor to numpy. Use Tensor.cpu to copy the tensor to host memory first

6

Hello!
I’m delighted that the issue has been resolved! Your update with the outcome is greatly appreciated.

Best Regards,
Raymond
choice advantages