Android runtime execution - Execution of method forward failed with status 0x12

[adonnini]

Hi,

After loading my model in my Android application, when I attempted to run it execution failed with Execution of method forward failed with status 0x12 error.

Below you will find the trceback log produced by the error and the code where the error occurred.

I tried to look into the EValue and NativePeer source code but did not see anything that would explain what happened as I think the error occurred in NativePeer native source code (not visible to me)

Please let me know if you need any additional information

Thanks

SOURCE CODE

			float[] flat = flatten(tmpData);
			final long[]  shapeArrDataPytorchFlattened = new long[]{tmpData.length, 4, 1};
			arrDataPytorch = Tensor.fromBlob(flat, shapeArrDataPytorchFlattened);
			
			//arrDataPytorch IS float[][]
.
.
.

			try {
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - About to load module --- ");
				mModule = Module.load(moduleFileAbsoluteFilePath);   //EXECUTORCH
			} catch (Exception e) {
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - Loading of module FAILED --- ");
				throw new RuntimeException(e);
			}

			Tensor outputTensor = null;

			float[] scores = new float[1];

			try {
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - Abut to run inference --- ");
				outputTensor = mModule.forward()[0].toTensor();   //EXECUTORCH
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - outputTensor - "+outputTensor);

//FAILURE OCCURS HERE		outputTensor = mModule.forward(EValue.from(arrDataPytorch))[0].toTensor();   //EXECUTORCH
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - outputTensor - "+outputTensor);

			} catch (Exception e) {
				Log.i(TAG, " - neuralNetworkloadAndRunPytorch - Inference FAILED --- ");
				throw new RuntimeException(e);
			}

TRACEBACK LOG

10-04 12:28:02.544: E/ExecuTorch(21206): input 1 is none
10-04 12:28:02.546: I/NeuralNetworkService(21206):  - neuralNetworkloadAndRunPytorch - Inference FAILED --- 
10-04 12:28:02.547: W/System.err(21206): java.lang.RuntimeException: java.lang.Exception: Execution of method forward failed with status 0x12
10-04 12:28:02.547: W/System.err(21206): 	at com.android.contextq.neuralNetwork.NeuralNetworkService.neuralNetworkloadAndRunPytorch(NeuralNetworkService.java:3043)
10-04 12:28:02.547: W/System.err(21206): 	at com.android.contextq.neuralNetwork.NeuralNetworkService$NeuralNetworkServiceRunnable.run(NeuralNetworkService.java:10035)
10-04 12:28:02.548: W/System.err(21206): 	at java.lang.Thread.run(Thread.java:1119)
10-04 12:28:02.548: W/System.err(21206): Caused by: java.lang.Exception: Execution of method forward failed with status 0x12
10-04 12:28:02.548: W/System.err(21206): 	at org.pytorch.executorch.NativePeer.forward(Native Method)
10-04 12:28:02.548: W/System.err(21206): 	at org.pytorch.executorch.Module.forward(Module.java:75)
10-04 12:28:02.548: W/System.err(21206): 	at com.android.contextq.neuralNetwork.NeuralNetworkService.neuralNetworkloadAndRunPytorch(NeuralNetworkService.java:3035)
10-04 12:28:02.548: W/System.err(21206): 	... 2 more

cc @kirklandsign @cbilgin

[psiddh]

Hi @adonnini

Thanks for reporting this issue. Error 0x12 indicates InvalidArgument, and the "input 1 is none" message suggests your model expects multiple inputs but you're only providing one. (If this is indeed the case, we will improve the error logs to better reflect the underlying mismatch args)

To help debug this quickly, can you please provide:

1. Model input signature: How many inputs does your model expect? What are their shapes/dtypes?

2. Detailed logs: Please add this code and share the output:

   
   try {
       outputTensor = mModule.forward(EValue.from(arrDataPytorch))[0].toTensor();
   } catch (Exception e) {
       String logs = mModule.readLogBuffer();
       Log.e("ExecuTorch", "Detailed logs: " + logs, e);
       throw e;
   }
   
   

The readLogBuffer() output may show the expected input count, shapes, and types that your model needs.

3. Also please run: adb logcat | grep -i executorch
   and share the complete output, not just the exception, for full context

[adonnini]

@psiddh when you ask for the number of inputs what do oyu mean exactly?
here is the exported program creation line:

        exported_program = export(m, (enc_input, dec_input, dec_source_mask, dec_target_mask), strict=False)

Is this what you meant?
In the Android application, there is no call/need for the above inputs as I am running the model for inference not training it as you can see for the code snippet I sent when I opened this issue.

Thanks

[adonnini]

@psiddh here is the output from

Log.e("ExecuTorch", "Detailed logs: " + logs, e);

10-06 17:11:57.932: E/ExecuTorch(20383): Detailed logs: [Ljava.lang.String;@ec36093
10-06 17:11:57.932: E/ExecuTorch(20383): java.lang.Exception: Execution of method forward failed with status 0x12
10-06 17:11:57.932: E/ExecuTorch(20383): 	at org.pytorch.executorch.NativePeer.forward(Native Method)
10-06 17:11:57.932: E/ExecuTorch(20383): 	at org.pytorch.executorch.Module.forward(Module.java:75)
10-06 17:11:57.932: E/ExecuTorch(20383): 	at com.android.contextq.neuralNetwork.NeuralNetworkService.neuralNetworkloadAndRunPytorch(NeuralNetworkService.java:3036)
10-06 17:11:57.932: E/ExecuTorch(20383): 	at com.android.contextq.neuralNetwork.NeuralNetworkService$NeuralNetworkServiceRunnable.run(NeuralNetworkService.java:10050)
10-06 17:11:57.932: E/ExecuTorch(20383): 	at java.lang.Thread.run(Thread.java:1119)
10-06 17:11:57.932: I/NeuralNetworkService(20383):  - neuralNetworkloadAndRunPytorch - Inference FAILED --- 
10-06 17:11:57.933: W/System.err(20383): java.lang.RuntimeException: java.lang.Exception: Execution of method forward failed with status 0x12
10-06 17:11:57.933: W/System.err(20383): 	at com.android.contextq.neuralNetwork.NeuralNetworkService.neuralNetworkloadAndRunPytorch(NeuralNetworkService.java:3051)
10-06 17:11:57.933: W/System.err(20383): 	at com.android.contextq.neuralNetwork.NeuralNetworkService$NeuralNetworkServiceRunnable.run(NeuralNetworkService.java:10050)
10-06 17:11:57.933: W/System.err(20383): 	at java.lang.Thread.run(Thread.java:1119)
10-06 17:11:57.933: W/System.err(20383): Caused by: java.lang.Exception: Execution of method forward failed with status 0x12
10-06 17:11:57.934: W/System.err(20383): 	at org.pytorch.executorch.NativePeer.forward(Native Method)
10-06 17:11:57.934: W/System.err(20383): 	at org.pytorch.executorch.Module.forward(Module.java:75)
10-06 17:11:57.934: W/System.err(20383): 	at com.android.contextq.neuralNetwork.NeuralNetworkService.neuralNetworkloadAndRunPytorch(NeuralNetworkService.java:3036)
10-06 17:11:57.934: W/System.err(20383): 	... 2 more

[larryliu0820]

Hi @adonnini you can inspect the method_meta field from java API: https://docs.pytorch.org/executorch/main/javadoc/org/pytorch/executorch/MethodMetadata.html. There you can see how many input metadata

[larryliu0820]

cc @kirklandsign