update lac and seq2seq to api_1.8 (#4604)

* update lac and seq2seq to api_1.8

* update lac and seq2seq to api_1.8

* update lac and seq2seq to api_1.8

PaddleNLP/lexical_analysis/creator.py

@@ -105,15 +105,15 @@ def create_pyreader(args,
         # create lac pyreader
         if mode == 'train':
             pyreader.set_sample_list_generator(
-                fluid.io.batch(
-                    fluid.io.shuffle(
+                paddle.batch(
+                    paddle.reader.shuffle(
                         reader.file_reader(file_name),
                         buf_size=args.traindata_shuffle_buffer),
                     batch_size=args.batch_size / device_count),
                 places=place)
         else:
             pyreader.set_sample_list_generator(
-                fluid.io.batch(
+                paddle.batch(
                     reader.file_reader(
                         file_name, mode=mode),
                     batch_size=args.batch_size / device_count),

PaddleNLP/lexical_analysis/run_ernie_sequence_labeling.py

@@ -117,10 +117,10 @@ def do_train(args):
                 model="ernie",
                 place=place)
             
+            clip = fluid.clip.GradientClipByGlobalNorm(clip_norm=1.0)
             optimizer = fluid.optimizer.Adam(
-                learning_rate=args.base_learning_rate)
-            fluid.clip.set_gradient_clip(
-                clip=fluid.clip.GradientClipByGlobalNorm(clip_norm=1.0))
+                learning_rate=args.base_learning_rate, 
+                grad_clip=clip)
             optimizer.minimize(train_ret["avg_cost"])
 
     lower_mem, upper_mem, unit = fluid.contrib.memory_usage(

PaddleNLP/seq2seq/seq2seq/train.py

@@ -97,14 +97,14 @@ def main():
                 dropout=dropout)
         loss = model.build_graph()
         inference_program = train_program.clone(for_test=True)
-        fluid.clip.set_gradient_clip(clip=fluid.clip.GradientClipByGlobalNorm(
-            clip_norm=max_grad_norm))
+        clip=fluid.clip.GradientClipByGlobalNorm(
+            clip_norm=max_grad_norm)
         lr = args.learning_rate
         opt_type = args.optimizer
         if opt_type == "sgd":
-            optimizer = fluid.optimizer.SGD(lr)
+            optimizer = fluid.optimizer.SGD(lr, grad_clip=clip)
         elif opt_type == "adam":
-            optimizer = fluid.optimizer.Adam(lr)
+            optimizer = fluid.optimizer.Adam(lr, grad_clip=clip)
         else:
             print("only support [sgd|adam]")
             raise Exception("opt type not support")

PaddleNLP/seq2seq/variational_seq2seq/model.py

@@ -229,7 +229,7 @@ class VAE(object):
 
         # `sample_output_layer` samples an id from the logits distribution instead of argmax(logits)
         # it will be used within BeamSearchDecoder
-        sample_output_layer = lambda x: layers.unsqueeze(layers.one_hot(
+        sample_output_layer = lambda x: layers.unsqueeze(fluid.one_hot(
             layers.unsqueeze(
             layers.sampling_id(
                 layers.softmax(

PaddleNLP/seq2seq/variational_seq2seq/train.py

@@ -89,9 +89,8 @@ def main():
             inference_program = fluid.default_main_program().clone(
                 for_test=True)
 
-            fluid.clip.set_gradient_clip(
             clip=fluid.clip.GradientClipByGlobalNorm(
-                    clip_norm=max_grad_norm))
+                    clip_norm=max_grad_norm)
 
             learning_rate = fluid.layers.create_global_var(
                 name="learning_rate",
@@ -102,9 +101,9 @@ def main():
 
             opt_type = args.optimizer
             if opt_type == "sgd":
-                optimizer = fluid.optimizer.SGD(learning_rate)
+                optimizer = fluid.optimizer.SGD(learning_rate, grad_clip=clip)
             elif opt_type == "adam":
-                optimizer = fluid.optimizer.Adam(learning_rate)
+                optimizer = fluid.optimizer.Adam(learning_rate, grad_clip=clip)
             else:
                 print("only support [sgd|adam]")
                 raise Exception("opt type not support")
@@ -272,7 +271,7 @@ def main():
                           (old_lr, new_lr))
 
                     dir_name = args.model_path + "/epoch_" + str(best_epoch_id)
-                    fluid.io.load_params(exe, dir_name)
+                    fluid.load(main_program, dir_name, exe)
 
                     decay_cnt += 1
                     if decay_cnt == max_decay: