-
Notifications
You must be signed in to change notification settings - Fork 3.6k
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
M1 arm64 release binaries #3129
Comments
I managed to get it working on Apple M1 by installing protobuf from brew and then compiling onnx as suggested about. if protobuf is installed from pip then onnx compilation failed on my side
|
Some users will need:
I found 1.10.1 works for me but not 1.8.1 |
|
@islavutin I think I was on Py 3.8.2. I don't suppose adding the --upgrade helps you? Life on planet M1 is tricky. |
I can confirm that
For me the following worked out of the box:
I obtained Python |
Thanks @Fohlen I'll try that. |
@Fohlen the issue isn't with installing. If you run
|
I can import |
@Fohlen via Homebrew install. Are you running the arm64 version of
|
My python resolves to:
Which is Python 3.9.7 installed by PyEnv (without Rosetta). Maybe you can give that a try and see if it helps. The homebrew-ish version of Python has been known to cause many different issues down the road. |
@Fohlen same issue. Can you check the Python binary is
|
This is getting more and more strange, because:
So I have a working installation of |
@lutzroeder I found the following wheel in my cache: I double-checked if at some point I compiled and installed the wheel myself, but I did not. Let's dig into where my PyPi got that wheel from. |
Okay @lutzroeder the wheel is actually built from source. Please try the following:
Your prompt should look a little something like this:
If that still does not yield any results try installing with:
This will force-build the wheel from source using the legacy build system. |
@Fohlen first option shows the same error, second option a build error. For context, the issue here isn't to find a workaround. It is understood that build from source in a virtual environment works. This issue is tracking "regular" |
Well in my environment it doesn't, as from the pip log you can see that out of the box it builds an arm with PEP 517. This is the expected and correct behaviour, in case the wheel offers a build from source option. It is also the correct approach to packaging, since it's unrealistic to expect that every project could easily add an M1 to their build pipeline. |
The error when running
|
Hello, I need
I get this error:
can someone help me out with this? Is it possible at the moment to get onnxruntime running on M1 Macs? |
@Bresenham try:
|
@Fohlen Thank you, but same problem. I guess the problem is Is it enough to Edit: Okay, after |
Maybe a bit late but in case someone stumbles over this thread: In a Miniforge Arm environment set up with the instructions from Apple to install Tensorflow, I have installed onnx using |
|
In a similar manner i found 1.12.0 works for me |
pip install onnx doesn't work in some M1 silicon Mac, and installing it with conda is supposed to be a solution onnx/onnx#3129 IDK but it is somehow related with protobuf version too onnx/onnx#4475
There's an issue that onnx, dependency of invisible-watermark, is not properly installed on the M1 silicons. And it is still opened onnx/onnx#3129 Therefore we should install onnx with conda instead of pip in prior to installing with pip. Otherwise other dependencies would be skipped without being installed.
Seriously haven't fixed this yet? |
For me it was:
Works for 1.12.0.
|
@apetsekin what version of python are you using? |
Bug Report
Describe the bug
ONNX fails to install on Apple M1.
System information
Reproduction instructions
The text was updated successfully, but these errors were encountered: