Disable per-channel quant in requantize_flatbuffer_test

Author: rascani

tensorflow/lite/micro/tools/requantize_flatbuffer_test.py

@@ -24,7 +24,7 @@
 from tflite_micro.tensorflow.lite.tools import flatbuffer_utils
 
 
-#TODO(b/248061370): replace the keras model creation process with flatbuffer manipulation to speed up test
+# TODO(b/248061370): replace the keras model creation process with flatbuffer manipulation to speed up test
 def create_simple_fc_model():
   '''Create a simple model with two fully connected(fc) layers'''
   model = tf.keras.models.Sequential([
@@ -60,6 +60,8 @@ def convert_tfl_converter(keras_model,
         EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8
     ]
   converter.representative_dataset = representative_dataset_gen
+  # TODO(b/324385802): Support per-channel quantization for FullyConnected.
+  converter._experimental_disable_per_channel_quantization_for_dense_layers = True
   return converter.convert()