AnalysisQAT fit new Quant (#1630)

paddleslim/quant/analysis_qat.py

@@ -160,7 +160,7 @@ class AnalysisQAT(object):
                                                   op_node.input('X')[0])
                 out_var = graph._find_node_by_name(op_node.outputs,
                                                    op_node.output('Y')[0])
-                if 'quantized' in in_var.name():
+                if not in_var.persistable():
                     # act
                     for op in graph.all_op_nodes():
                         o_ns = op.output_arg_names()
@@ -173,8 +173,9 @@ class AnalysisQAT(object):
                 else:
                     # weight
                     with paddle.static.scope_guard(float_scope):
+                        float_name = in_var.name().replace('.quantized', '')
                         float_weight = np.array(
-                            float_scope.find_var(in_var.name()).get_tensor())
+                            float_scope.find_var(float_name).get_tensor())
                     with paddle.static.scope_guard(quant_scope):
                         quant_scope.find_var(in_var.name()).get_tensor().set(
                             float_weight, self.places)
@@ -216,7 +217,7 @@ class AnalysisQAT(object):
             with paddle.static.scope_guard(float_scope):
                 float_preds = executor.run(program=float_program,
                                            feed=data,
-                                           fetch_list=self.fetch_list,
+                                           fetch_list=self.float_fetch_list,
                                            return_numpy=False)
                 float_preds = float_preds[0]
             with paddle.static.scope_guard(quant_scope):
@@ -253,11 +254,12 @@ class AnalysisQAT(object):
                     idx + 1, len(self.inputs_of_quantized_op), weight_name))
 
             with paddle.static.scope_guard(float_scope):
-                [float_program, _, _] = load_inference_model(
-                    self.float_model_dir,
-                    executor=executor,
-                    model_filename=self.model_filename,
-                    params_filename=self.params_filename)
+                [float_program, self.float_feed_list,
+                 self.float_fetch_list] = load_inference_model(
+                     self.float_model_dir,
+                     executor=executor,
+                     model_filename=self.model_filename,
+                     params_filename=self.params_filename)
 
             with paddle.static.scope_guard(quant_scope):
                 [program, self.feed_list,