TFLITE model

[OlSv]

Hello, community,

Few days ago, I created a new tflite model with Vertex AI. When my app tries to use the model for Object Detection, the application crashes. The same code with the older tflite model works OK. 

The difference in Logcat between the old and new versions is a line in bold (quote from Logcat below). "Replacing nodes..." appears for the run with the new tflite model. When detectobject is called, "Fatal signal 6 (SIGABRT),..." starts crash proceedings.

Please, help to fix the problem

tflite Initialized TensorFlow Lite runtime.
tflite Created TensorFlow Lite XNNPACK delegate for CPU.
tflite Replacing 16 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 17 partitions.

.....

libc ....Fatal signal 6 (SIGABRT), code -1 (SI_QUEUE) in tid 5792 (pool-18-thread-), pid 5589 (...)

[MJane]

Hi @OlSv , 

Welcome to Google Cloud Community!

The log shows that your TensorFlow Lite app ran into an error, specifically signal 6 (SIGABRT), which usually means the program stopped because it hit a problem it couldn't handle.

Here are workarounds that you might consider to help you check the cause and possible solution of the issue: 

• Update Dependencies -  Ensure that you are using the latest version of TensorFlow Lite and related libraries, as there may be bug fixes or improvements. You can read through this documentation for the latest updates on TensorFlow Lite now known as LiteRT. 
• Review the Code - Look for any potential issues in your code that might be causing the crash.
• Migrate to Vertex AI - Ensure to migrate your application to Vertex AI if you are using Legacy AutoML. As per documentation, the option to train or update models for text classification, entity extraction, and sentiment analysis in Vertex AI AutoML will no longer be supported.
• Test Without Delegates - Try running your model without the XNNPACK delegate to see if the issue persists. This might help identify if the problem is related to the delegate.

If the error persists, I suggest contacting Google Cloud Support as they can provide more insights to see if the behavior you've encountered is a known issue or specific to your project.

I hope the above information is helpful.