Mac M1 pro segmentation fault on import

[axmav]

🐛 Describe the bug

Hello! Can not import torchvision on my mac. Get an error:
libc++abi: terminating with uncaught exception of type std::length_error: vector
as well as segmentation fault.

code:

import torch
from torch import nn
from torch.nn import functional as F
from torch.distributions.normal import Normal
from torch.distributions.kl import kl_divergence
from torchvision.utils import save_image

it also fails just with from torchvision.utils import save_image

Similar issue: pyg-team/pytorch_geometric#4419
But not a single proposed solution works...

Versions

PyTorch version: 1.10.2
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 12.1 (arm64)
GCC version: Could not collect
Clang version: 13.1.6 (clang-1316.0.21.2.3)
CMake version: Could not collect
Libc version: N/A

Python version: 3.9.10 | packaged by conda-forge | (main, Feb 1 2022, 21:25:34) [Clang 11.1.0 ] (64-bit runtime)
Python platform: macOS-12.1-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Versions of relevant libraries:
[pip3] numpy==1.21.5
[pip3] torch==1.10.2
[pip3] torchaudio==0.11.0
[pip3] torchvision==0.12.0
[conda] numpy 1.21.5 py39h47fb9ce_0 conda-forge
[conda] torch 1.10.2 pypi_0 pypi
[conda] torchaudio 0.11.0 pypi_0 pypi
[conda] torchvision 0.12.0 pypi_0 pypi

[datumbox]

TorchVision doesn't properly provide support for this at the moment. We have a todo to investigate at #5171 but we don't currently have the bandwidth to resolve.

We'll definitely need to fix once when we investigate. Also by briefly checking the linked issue but I'm not sure what exactly it's invoking the libc++abi when you simply import from torchvision.utils import save_image. I wonder if that's linked to the @torch.no_grad() decorator. Could you test this out?

[axmav]

@datumbox could you provide me with a code? Idk what is decorator.
But I solved this problem by downgrading to pip install "torchvision==0.11.3"

[datumbox]

@axmav I was wonder what would happen if you try to import a dummy method like this:

import torch

@torch.no_grad()
def dummy():
    return 0

[axmav]

@axmav I was wonder what would happen if you try to import a dummy method like this:

import torch

@torch.no_grad()
def dummy():
    return 0

yes, this works on 0.12.0

[finnatsea]

I also had this issue on m1. Fixed by using pip install "torchvision==0.11.3"

[liderrick616]

doesn't work, still get segmentation 11, on Mac M2