clean hapi notes, clean eager_run and modify num_labels to num_classes (#5045)

PaddleNLP/examples/glue/run_glue.py

@@ -150,8 +150,6 @@ def parse_args():
     )
     parser.add_argument(
         "--seed", default=42, type=int, help="random seed for initialization")
-    parser.add_argument(
-        "--eager_run", default=True, type=eval, help="Use dygraph mode.")
     parser.add_argument(
         "--n_gpu",
         default=1,
@@ -255,7 +253,6 @@ def convert_example(example,
 
 
 def do_train(args):
-    paddle.enable_static() if not args.eager_run else None
     paddle.set_device("gpu" if args.n_gpu else "cpu")
     if paddle.distributed.get_world_size() > 1:
         paddle.distributed.init_parallel_env()
@@ -324,10 +321,10 @@ def do_train(args):
             num_workers=0,
             return_list=True)
 
-    num_labels = 1 if train_dataset.get_labels() == None else len(
+    num_classes = 1 if train_dataset.get_labels() == None else len(
         train_dataset.get_labels())
     model = model_class.from_pretrained(
-        args.model_name_or_path, num_labels=num_labels)
+        args.model_name_or_path, num_classes=num_classes)
     if paddle.distributed.get_world_size() > 1:
         model = paddle.DataParallel(model)
 
@@ -362,15 +359,6 @@ def do_train(args):
 
     metric = metric_class()
 
-    ### TODO: use hapi
-    # trainer = paddle.hapi.Model(model)
-    # trainer.prepare(optimizer, loss_fct, paddle.metric.Accuracy())
-    # trainer.fit(train_data_loader,
-    #             dev_data_loader,
-    #             log_freq=args.logging_steps,
-    #             epochs=args.num_train_epochs,
-    #             save_dir=args.output_dir)
-
     global_step = 0
     tic_train = time.time()
     for epoch in range(args.num_train_epochs):

PaddleNLP/paddlenlp/transformers/electra/modeling.py

@@ -436,11 +436,11 @@ class ElectraGenerator(ElectraPretrainedModel):
 class ElectraClassificationHead(nn.Layer):
     """Head for sentence-level classification tasks."""
 
-    def __init__(self, hidden_size, hidden_dropout_prob, num_labels):
+    def __init__(self, hidden_size, hidden_dropout_prob, num_classes):
         super(ElectraClassificationHead, self).__init__()
         self.dense = nn.Linear(hidden_size, hidden_size)
         self.dropout = nn.Dropout(hidden_dropout_prob)
-        self.out_proj = nn.Linear(hidden_size, num_labels)
+        self.out_proj = nn.Linear(hidden_size, num_classes)
 
     def forward(self, features, **kwargs):
         x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
@@ -453,13 +453,13 @@ class ElectraClassificationHead(nn.Layer):
 
 
 class ElectraForSequenceClassification(ElectraPretrainedModel):
-    def __init__(self, electra, num_labels):
+    def __init__(self, electra, num_classes):
         super(ElectraForSequenceClassification, self).__init__()
-        self.num_labels = num_labels
+        self.num_classes = num_classes
         self.electra = electra
         self.classifier = ElectraClassificationHead(
             self.electra.config["hidden_size"],
-            self.electra.config["hidden_dropout_prob"], self.num_labels)
+            self.electra.config["hidden_dropout_prob"], self.num_classes)
 
         self.init_weights()
 
@@ -478,13 +478,13 @@ class ElectraForSequenceClassification(ElectraPretrainedModel):
 
 
 class ElectraForTokenClassification(ElectraPretrainedModel):
-    def __init__(self, electra, num_labels):
+    def __init__(self, electra, num_classes):
         super(ElectraForTokenClassification, self).__init__()
-        self.num_labels = num_labels
+        self.num_classes = num_classes
         self.electra = electra
         self.dropout = nn.Dropout(self.electra.config["hidden_dropout_prob"])
         self.classifier = nn.Linear(self.electra.config["hidden_size"],
-                                    self.num_labels)
+                                    self.num_classes)
         self.init_weights()
 
     def forward(self,