RuntimeError On Mac M1 Chip

mkhoshle commented 2 years ago

Hi, I run your code on my Mac M1 chip after installing Pytorch and verifying that. How ever I get the following error:

(ml) Mahzads-MacBook-Pro-2:pytorch-m1-gpu mahzadkhoshlessan$ python lenet-mnist.py --device "mps"
Traceback (most recent call last):
  File "/Users/mahzadkhoshlessan/Desktop/machine-learning-notes/benchmark/pytorch-m1-gpu/lenet-mnist.py", line 253, in <module>
    DEVICE = torch.device(args.device)
RuntimeError: Expected one of cpu, cuda, xpu, mkldnn, opengl, opencl, ideep, hip, ve, ort, mlc, xla, lazy, vulkan, meta, hpu device type at start of device string: mps

Any idea?

rasbt commented 2 years ago

did you install the nightly release? The M1 GPU support is not in the main release yet. You can do that by

pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu

mkhoshle commented 2 years ago

I installed using conda install pytorch torchvision -c pytorch given instruction in the page: https://pytorch.org/get-started/locally/. Now I get this error:

INTEL MKL ERROR: dlopen(/opt/miniconda3/envs/ml/lib/libmkl_intel_thread.1.dylib, 0x0009): Library not loaded: @rpath/libiomp5.dylib
  Referenced from: /opt/miniconda3/envs/ml/lib/libmkl_intel_thread.1.dylib
  Reason: tried: '/opt/miniconda3/envs/ml/lib/libiomp5.dylib' (no such file), '/opt/miniconda3/envs/ml/bin/../lib/libiomp5.dylib' (no such file), '/usr/local/lib/libiomp5.dylib' (no such file), '/usr/lib/libiomp5.dylib' (no such file).
Intel MKL FATAL ERROR: Cannot load libmkl_intel_thread.1.dylib.

rasbt commented 2 years ago

If you want to install the nightly version, I think you have to use the -c pytorch-nightly channel as they recommend here: https://pytorch.org

mkhoshle commented 2 years ago

I see. Ok I uninstalled and reinstalled using conda install pytorch torchvision torchaudio -c pytorch-nightly. But now I get this message:

(ml) Mahzads-MacBook-Pro-2:pytorch-m1-gpu mahzadkhoshlessan$ python lenet-mnist.py --device mps
OMP: Error #15: Initializing libiomp5.dylib, but found libomp.dylib already initialized.
OMP: Hint This means that multiple copies of the OpenMP runtime have been linked into the program. That is dangerous, since it can degrade performance or cause incorrect results. The best thing to do is to ensure that only a single OpenMP runtime is linked into the process, e.g. by avoiding static linking of the OpenMP runtime in any library. As an unsafe, unsupported, undocumented workaround you can set the environment variable KMP_DUPLICATE_LIB_OK=TRUE to allow the program to continue to execute, but that may cause crashes or silently produce incorrect results. For more information, please see http://www.intel.com/software/products/support/.
Abort trap: 6

rasbt commented 2 years ago

Huh that's interesting. It's weird that it mentions Intel in this message

For more information, please see http://www.intel.com/software/products/support/.

You have a computer with M1 chip though, right?

mkhoshle commented 2 years ago

Correct! and I am not sure either!

rasbt commented 2 years ago

could you maybe try this in a fresh environment and see if that helps:

$conda create -n torch-nightly python=3.8 

$ conda activate torch-nightly

$ pip install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu

mkhoshle commented 2 years ago

I get the following error:

(torch-nightly) Mahzads-MacBook-Pro-2:pytorch-m1-gpu mahzadkhoshlessan$ python lenet-mnist.py --device mps
torch 1.12.0.dev20220520
device mps
Traceback (most recent call last):
  File "lenet-mnist.py", line 286, in <module>
    model = model.to(DEVICE)
  File "/opt/miniconda3/envs/torch-nightly/lib/python3.8/site-packages/torch/nn/modules/module.py", line 927, in to
    return self._apply(convert)
  File "/opt/miniconda3/envs/torch-nightly/lib/python3.8/site-packages/torch/nn/modules/module.py", line 579, in _apply
    module._apply(fn)
  File "/opt/miniconda3/envs/torch-nightly/lib/python3.8/site-packages/torch/nn/modules/module.py", line 579, in _apply
    module._apply(fn)
  File "/opt/miniconda3/envs/torch-nightly/lib/python3.8/site-packages/torch/nn/modules/module.py", line 602, in _apply
    param_applied = fn(param)
  File "/opt/miniconda3/envs/torch-nightly/lib/python3.8/site-packages/torch/nn/modules/module.py", line 925, in convert
    return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
RuntimeError: PyTorch is not linked with support for mps devices

rasbt commented 2 years ago

Hmm ... could you run the following code and show the output?

conda install watermark -c conda-forge
python -c "import watermark; print(watermark.watermark())"

mkhoshle commented 2 years ago

Yes here is the output:

(torch-nightly) Mahzads-MacBook-Pro-2:pytorch-m1-gpu mahzadkhoshlessan$ python -c "import watermark; print(watermark.watermark())" 
Last updated: 2022-05-20T15:06:56.465751-04:00

Python implementation: CPython
Python version       : 3.8.13
IPython version      : 8.3.0

Compiler    : Clang 12.0.0 
OS          : Darwin
Release     : 21.4.0
Machine     : x86_64
Processor   : i386
CPU cores   : 8
Architecture: 64bit

rasbt commented 2 years ago

Processor : i386

Oh, looks like you have an Intel Mac, not Arm Mac

mkhoshle commented 2 years ago

Ok but then what does this mean? What will change when installing Pytorch on my laptop?

rasbt commented 2 years ago

You can maybe double check with your system configuration info, but it doesn't look like your computer has a M1 chip. So that means that you can just keep using the regular CPU version for PyTorch, but the "mps" option is not going to work for you.

Screen Shot 2022-05-20 at 2 36 59 PM