Dev adding fln actq holder init in model (#25)

* add test code for FLN

* modified the model

* add space

* minor corrections

* fix fw_info in test code

* add FLN_NO_QUANT test

* change 16bit for FLN

* Resolve test errors

* Revert "Resolve test errors"

This reverts commit 0e77645.

* Support for QuantizationConfig changes

* Added an argument to the GradientPTQConfig function.

* Change description format

* Revert "Change description format"

This reverts commit c83f4a4.

---------

Co-authored-by: kawakami-masaki0 <kawakami.masaki@jp.panasonic.com>

Author: KazunoriSumiya

tests/external_tests/keras_tests/models_tests/test_networks_runner.py

@@ -97,15 +97,21 @@ def compare(self, inputs_list, quantized_model, qc, tpc):
                 self.unit_test.assertTrue(False, f'fail TFLite convertion with the following error: {error_msg}')
 
     def run_network(self, inputs_list, qc, tpc):
+        from model_compression_toolkit.gptq.common.gptq_config import GradualActivationQuantizationConfig
         def representative_data_gen():
             for _ in range(self.num_calibration_iter):
                 yield inputs_list
 
         core_config = mct.core.CoreConfig(quantization_config=qc)
         if self.gptq:
-            arc = mct.gptq.GradientPTQConfig(n_epochs=2, optimizer=tf.keras.optimizers.Adam(
-                learning_rate=0.0001), optimizer_rest=tf.keras.optimizers.Adam(
-                learning_rate=0.0001), loss=multiple_tensors_mse_loss)
+            arc = mct.gptq.GradientPTQConfig(n_epochs=2,
+                                             optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
+                                             optimizer_rest=tf.keras.optimizers.Adam(learning_rate=0.0001),
+                                             loss=multiple_tensors_mse_loss,
+                                             train_bias=True,
+                                             hessian_weights_config=None,
+                                             gradual_activation_quantization_config= GradualActivationQuantizationConfig(),
+                                             regularization_factor=1)
 
             ptq_model, quantization_info = mct.gptq.keras_gradient_post_training_quantization(
                 self.model_float,

tests_pytest/pytorch_tests/unit_tests/core/back2framework/test_pytorch_model_builder_fln.py

@@ -59,7 +59,6 @@ def build_qc(q_mode=ActivationQuantizationMode.QUANT):
         signedness=Signedness.AUTO
     )
     a_qcfg = NodeActivationQuantizationConfig(op_cfg=op_cfg)
-    a_qcfg.set_qc(QuantizationConfig())
     a_qcfg.quant_mode = q_mode
     w_qcfg = NodeWeightsQuantizationConfig(op_cfg=op_cfg,
                                            weights_channels_axis=ChannelAxisMapping(0, 1),