My .tflite model works on python but it dosen’t work well on android project.

It seems no difference between python and android.
What kinds of layer or function changes demention of input?

Standalone code to reproduce the issue

model
 python
model = Sequential()
model.add(Input([256], dtype="int32"))
model.add(Embedding(35000, 10))
model.add(GRU(10))
# model.add(Dropout(0.5))
model.add(Dense(2, activation='softmax', ))
model.compile(loss='binary_crossentropy', optimizer="adam", metrics=['accuracy'])
print(model.summary())

Model: “sequential_43”
┏━━━━━━━━━━━━┳━━━━━━━━━┳━━━━━━┓
┃ Layer (type) ┃ Output Shape ┃ Param # ┃
┡━━━━━━━━━━━━╇━━━━━━━━━╇━━━━━━┩
│ embedding_35 (Embedding)│ (None, 256, 10) │ 350,000 │
├──────────────────┼─────────────┼─────────┤
│ gru_33 (GRU) │ (None, 10) │ 660 │
├──────────────────┼─────────────┼─────────┤
│ dense_27 (Dense) │ (None, 2) │ 22 │
└──────────────────┴─────────────┴─────────┘
Total params: 350,682 (1.34 MB)
Trainable params: 350,682 (1.34 MB)
Non-trainable params: 0 (0.00 B)

error occur

private AnalysisResult analyzeTextTFLite(String text) {
        // convert text to custom ids
        Feature feature = featureConverter.convert(text, ADD_SPECIAL_TOKENS);
        int curSeqLen = feature.inputIds.length;

        int[] tfInputs = new int[256];
        for (int j = 0; j < curSeqLen; j++) {
            tfInputs[j] = feature.inputIds[j];
        }

        Map<String, Object> inputsMap = new HashMap<>();
        Map<String, Object> outputMap = new HashMap<>();

        inputsMap.put("keras_tensor_198", tfInputs);

        float[][] logits = new float[1][2];
        outputMap.put("output_0", logits);

        final long moduleForwardStartTime = SystemClock.elapsedRealtime();
        tflite.runSignature(inputsMap, outputMap);
        Log.d(TAG, "Model inference score : " + logits[0][0] + "," + logits[0][1]);

        float[] scores = new float[2];
        scores[0] = argmax(logits[0]);

        final long moduleForwardDuration = SystemClock.elapsedRealtime() - moduleForwardStartTime;
        return new AnalysisResult(scores, moduleForwardDuration);
    }

works well

input_details = interpreter.get_input_details()
print(input_details)
"""
[{'name': 'serving_default_keras_tensor_198:0',
  'index': 0,
  'shape': array([  1, 256]),
  'shape_signature': array([ -1, 256]),
  'dtype': numpy.int32,
  'quantization': (0.0, 0),
  'quantization_parameters': {'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32),
   'quantized_dimension': 0},
  'sparsity_parameters': {}}]
output_details = interpreter.get_output_details()
print(output_details)
"""

output_data = interpreter.get_tensor(output_details[0]['index'])
print(output_data)
"""
[{'name': 'StatefulPartitionedCall_1:0',
  'index': 36,
  'shape': array([1, 2]),
  'shape_signature': array([-1,  2]),
  'dtype': numpy.float32,
  'quantization': (0.0, 0),
  'quantization_parameters': {'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32),
   'quantized_dimension': 0},
  'sparsity_parameters': {}}]
"""

interpreter.set_tensor(input_details[0]['index'], encode_plus_inputs["input_ids"])
interpreter.invoke()

Relevant log output

FATAL EXCEPTION: ModuleActivity
                                                                                                    Process: org.pytorch.demo, PID: 32712
                                                                                                    java.lang.IllegalArgumentException: Internal error: Failed to run on the given Interpreter: tensorflow/lite/kernels/transpose.cc:63 op_context->perm->dims->data[0] != dims (3 != 2)
                                                                                                    Node number 10 (TRANSPOSE) failed to prepare.
                                                                                                        at org.tensorflow.lite.NativeInterpreterWrapper.run(Native Method)
                                                                                                        at org.tensorflow.lite.NativeInterpreterWrapper.run(NativeInterpreterWrapper.java:264)
                                                                                                        at org.tensorflow.lite.NativeInterpreterWrapper.runSignature(NativeInterpreterWrapper.java:194)
                                                                                                        at org.tensorflow.lite.Interpreter.runSignature(Interpreter.java:271)
                                                                                                        at org.tensorflow.lite.Interpreter.runSignature(Interpreter.java:284)
                                                                                                        at org.pytorch.demo.nlp.NSMCPytorchActivity.analyzeTextKoElectraTFLite(NSMCPytorchActivity.java:281)
                                                                                                        at org.pytorch.demo.nlp.NSMCPytorchActivity.analyzeText(NSMCPytorchActivity.java:201)
                                                                                                        at org.pytorch.demo.nlp.NSMCPytorchActivity.lambda$new$2$org-pytorch-demo-nlp-NSMCPytorchActivity(NSMCPytorchActivity.java:154)
                                                                                                        at org.pytorch.demo.nlp.NSMCPytorchActivity$$ExternalSyntheticLambda2.run(Unknown Source:4)
                                                                                                        at android.os.Handler.handleCallback(Handler.java:958)
                                                                                                        at android.os.Handler.dispatchMessage(Handler.java:99)
                                                                                                        at android.os.Looper.loopOnce(Looper.java:230)
                                                                                                        at android.os.Looper.loop(Looper.java:319)
                                                                                                        at android.os.HandlerThread.run(HandlerThread.java:67)

I think my model works well in android code..