tensorflow: TFLite Interpreter, allocate_tensors() failed to prepare, not kTFLiteInt8/Uint8

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): v1.14.0-rc1-22-gaf24dc91b5 1.14.0
• Python version: 3.6.3
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

Describe the current behavior The graph only consist of tf.split(), where I pin down to for this issue. After quantization with representative_dataset(), the interpreter fail to allocate_tensor(). RuntimeError: tensorflow/lite/kernels/dequantize.cc:62 op_context.input->type == kTFLiteUInt8 || op_context.input->type == kTFLiteInt8 was not true.Node number 0 (DEQUANTIZE) failed to prepare. I assume tf.split() cannot be quantize, and nothing to be quantize actually, since if I add tf.lite.OpsSet.TFLITE_BUILTINS_INT8 to fully quantize into INT8, it tell me SPLIT_V is not supported. So this simple graph should not have any quantize/dequantize at tf.split ops? Thus checking type if is Int8 and has dequantize layer here seems not to make sense?

Describe the expected behavior Interpreter sucessfully

Code to reproduce the issue

inputs_raw = tf.placeholder(tf.float32, shape=[1, 32, 32, 1], name='inputs_raw')
outputs = tf.split(value = inputs_raw, num_or_size_splits = 2, axis = 1)[0]

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    converter = lite.TFLiteConverter.from_session(sess, [inputs_raw], [outputs])

    converter.optimizations = [lite.Optimize.DEFAULT]
    def representative_data_gen():
        for i in range(1):
            yield [np.random.random_sample((1, 32, 32, 1)).astype(np.float32)]
            #yield [sess.run(tf.random.normal(inputs_raw.get_shape(), dtype = tf.float32, seed = 1))] # both have issue
    converter.representative_dataset = representative_data_gen
    tflite_model = converter.convert()
    open("converted_model_quant_test.tflite", "wb").write(tflite_model)

interpreter = lite.Interpreter(model_path = "converted_model_quant_test.tflite")
nterpreter.allocate_tensors()

Other info / logs Traceback (most recent call last): File “main.py”, line 532, in <module> interpreter.allocate_tensors() File “/proj/gpu_xxx/.local/lib/python3.6/site-packages/tensorflow/lite/python/interpreter.py”, line 95, in allocate_tensors return self._interpreter.AllocateTensors() File “/proj/gpu_xxx/.local/lib/python3.6/site-packages/tensorflow/lite/python/interpreter_wrapper/tensorflow_wrap_interpreter_wrapper.py”, line 106, in AllocateTensors return _tensorflow_wrap_interpreter_wrapper.InterpreterWrapper_AllocateTensors(self) RuntimeError: tensorflow/lite/kernels/dequantize.cc:62 op_context.input->type == kTfLiteUInt8 || op_context.input->type == kTfLiteInt8 was not true.Node number 0 (DEQUANTIZE) failed to prepare.