User Input of the Dynamic-Shape Model Does Not Match the Model Inference Result

Symptom

Scenario 1: An error is reported in the model execution. The log contains the following key information:

[INFO] GE(2284607, ir_build):2021-11-08-19:31:33.920.888 [task_context.cc:257]2284607 AllocateOutput: To allocate output for node: add2. index = 0, tensor desc = [TensorDescl DataType = 6 , Format = 2 · Shape = [50, ] 
[ERROR] GE(2284607, ir_build):2021-11-08-19:31:33.920.900 [task_context.cc:257]2284607 AllocateOutput: ErrorNo: 50331649() [LOAD][LOAD][Check][Size] add2(Add) index[0] mem size out of range! Expected size: 128, but given input size: 2. 

Scenario 2: An error is reported in the single-operator scenario. The log contains the following key information:

[INFO] GE(130191,python):2021-09-27-17:23:17.922.047 [single op.cc:161]131511 ValidateArgs:lnput [1], aligned_size:96, inputs.length:40, input_sizes_:96 
[INFO] GE(130191,python):2021-09-27-17:23:17.922.054 [single op.cc:186]131511 ValidateArgs:Output [0], aligned_size:3145760, outputs.length:3145728, input_sizes_:18446744073709551615 
[ERROR] GE(130191,python):2021-09-27-17:23:17,922.095 [single op.cc:161]131511 ValidateArgs: ErrorNo: 145000(Parameter invalid.) [FINALI][FlNALl][CheCkl][Param.Outputs]Output Size mismatch. index = 0, model expect  18446744073709551615, but given ....
[ERROR] ASCENDCL(130191,python):2021-09-27-17:23:17,922.213 [op_executor.cpp:68]131511 DoExecteAsync:  [FINALI][FlNALl][Exec][Op]Execte op failed. ge result = 145000

Possible Cause

Scenario 1:

During model execution, the size of output_buffer allocated by the user does not match that output by the model after layers of infershape.

Scenario 2:

In the single-operator scenario, the size of output_buffer allocated by the user does not match output_size calculated by infershape during single-operator compilation.

Solution

To rectify the fault, perform the following steps:

For scenario 1:

If this error occurs, check whether the size of the allocated output_buffer is correct. The size must be the same as that inferred by the model. You are advised to allocate output_buffer based on the size specified in the reported error message.

For scenario 2:

If this error occurs, check whether the size of the allocated output_buffer is the same as size specified by infershape. You are advised to allocate the buffer based on the error message.