Couldn't load libjnidnnl.dylib #1518
Labels
Comments
|
Please set the "org.bytedeco.javacpp.logger.debug" system property to "true" to get more information on the console. |
Already did, part of the log is attached to the issue, here's the full log from the start: |
|
So it looks like libiomp5.dylib isn't getting extracted from the JAR file for some reason. The only possible explanation is that it's not in your class path. What are the files in the JAR files in your class path? |
|
Actually the problem seems to be that TensorFlow is trying to load libiomp5.dylib for some reason, but it's not bundled in its archive. Try to load DNNL first, that should do the trick. |
How do I do that exactly? |
|
I added this in main(): Loader.load(dnnl.class);And it looks like there are no issues when loading the lib. However, I get this: Full log: |
|
Right, the API changed a lot with version 3, you'll need to update your code for that |
|
That's an error with ONNX Runtime though. That method can be found in SessionOptionsImpl: |
|
You'll need to recompile your source code to be able to reroute the call to SessionOptionsImpl |
|
OK, thanks. So it's not possible to use onnxruntime 1.0.0-M2.1 with javacpp 1.5.10, correct? The thing is, I tried doing that because I use 1.5.10 elsewhere in my project and if don't change anything, I get this warning about opencv, dnnl and onnxruntime not matching (because it uses 1.5.7 instead of 1.5.10). What also worries me, is that after switching back to 1.5.7 for onnxruntime, the code runs fine on Intel Macs, but crashes on Apple Silicon (in Rosetta mode) for some reason: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Caused by this:
Debug log:
otool output:
The text was updated successfully, but these errors were encountered: