[BugFix] Fix amp usage for evaluation. (#3303)

* fix eval of amp usage.

* fix

Author: ZHUI

model_zoo/ernie-1.0/run_pretrain.py

@@ -217,33 +217,43 @@ def run_evaluate(data_loader,
     for eval_step, batch in enumerate(data_loader):
         input_ids, segment_ids, input_mask, masked_lm_positions, \
         masked_lm_labels, next_sentence_labels = batch
+        with paddle.amp.auto_cast(args.use_amp,
+                                  custom_white_list=[
+                                      'softmax',
+                                      'layer_norm',
+                                      'gelu',
+                                  ],
+                                  custom_black_list=[
+                                      "c_softmax_with_cross_entropy",
+                                  ],
+                                  level=args.fp16_opt_level):
 
-        if args.binary_head:
-            prediction_scores, seq_relationship_score = model(
-                input_ids=input_ids,
-                token_type_ids=segment_ids,
-                position_ids=None,
-                attention_mask=input_mask,
-                masked_positions=masked_lm_positions)
-
-            lm_loss, sop_loss = criterion(prediction_scores,
-                                          seq_relationship_score,
-                                          masked_lm_labels,
-                                          next_sentence_labels)
-            loss = lm_loss + sop_loss
-        else:
-            prediction_scores = model(input_ids=input_ids,
-                                      token_type_ids=segment_ids,
-                                      position_ids=None,
-                                      attention_mask=input_mask,
-                                      masked_positions=masked_lm_positions)
-
-            loss = criterion(prediction_scores, None, masked_lm_labels)
-
-        loss_global["loss"] += loss.detach()
-        if args.binary_head:
-            loss_global["lm_loss"] += lm_loss.detach()
-            loss_global["sop_loss"] += sop_loss.detach()
+            if args.binary_head:
+                prediction_scores, seq_relationship_score = model(
+                    input_ids=input_ids,
+                    token_type_ids=segment_ids,
+                    position_ids=None,
+                    attention_mask=input_mask,
+                    masked_positions=masked_lm_positions)
+
+                lm_loss, sop_loss = criterion(prediction_scores,
+                                              seq_relationship_score,
+                                              masked_lm_labels,
+                                              next_sentence_labels)
+                loss = lm_loss + sop_loss
+            else:
+                prediction_scores = model(input_ids=input_ids,
+                                          token_type_ids=segment_ids,
+                                          position_ids=None,
+                                          attention_mask=input_mask,
+                                          masked_positions=masked_lm_positions)
+
+                loss = criterion(prediction_scores, None, masked_lm_labels)
+
+            loss_global["loss"] += loss.detach()
+            if args.binary_head:
+                loss_global["lm_loss"] += lm_loss.detach()
+                loss_global["sop_loss"] += sop_loss.detach()
 
         if eval_step >= iter_steps - 1:
             log_info_dict = dict()