-
Notifications
You must be signed in to change notification settings - Fork 197
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
Debug in .cc file. org.tensorflow.exceptions.TFInvalidArgumentException: indices = 1 is not in [0, 1) #189
Comments
If I recall correctly, I used to do that by
Then, gdb should break when your program reaches the native breakpoints |
It looks like your classpath might not be right as there are a bunch of JavaCPP errors at the top:
@saudet any idea how TF is managing to run if it can't load the JavaCPP pointer? |
It's just a warning, it still runs for backward compatibility, but it's obviously not recommended since users may hit loading errors... |
@uygnef , do you still want us to keep track of this issue, i.e. was the purpose only to know how to debug the TF C++ runtime from a Java process or you suspect that the out-of-range issue is related to the Java client? |
…ead of warnings (issue tensorflow/java#189)
@Craigacp I've changed the behavior in commit bytedeco/javacpp@f8ad06a so that those messages appear only in the debug log instead of as warnings. When someone encounters one of those corner cases, it'll still be easy to diagnose when that happens and then make sure they put |
@saudet What's the difference in behaviour when it can't load these JNI bindings? Is it slower or more resource intensive? If so then making what appears to be a packaging problem into an easily ignored logger message seems like a problem. Why don't we fix the fact that it can't load jnijavacpp by sorting out the packaging instead? |
It's not going to do anything differently unless it can't load those functions from a native library somewhere, so it's not a big issue. |
My projects which use TF-Java depend on what Maven tells me TF-Java depends on. So if the TF-Java pom files don't depend on a jar which contains jnijavacpp then we should fix the pom files. Given you wrote the JavaCPP related portions of TF-Java, I assumed we already had the necessary JavaCPP dependencies, both Java and native. If we don't have a complete set of them, then let's fix that. Which jar file is this binary in? |
For Linux, it's exactly this one: java/tensorflow-core/tensorflow-core-platform/pom.xml Lines 41 to 46 in 0268634
|
My tfserver project throw error when it perform an inference.
Something wrong when program get feature raw data and make feature to embedding tensor.
How can I debug into .cc file.
featue "cate_ord_start_hash" is created by
full error stack is
The text was updated successfully, but these errors were encountered: