Migrate metric to Evaluate in Pytorch examples (#18369)

* Migrate metric to Evaluate in pytorch examples

* Remove unused imports

examples/pytorch/question-answering/run_seq2seq_qa.py

@@ -25,8 +25,9 @@ from dataclasses import dataclass, field
 from typing import List, Optional, Tuple
 
 import datasets
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from trainer_seq2seq_qa import QuestionAnsweringSeq2SeqTrainer
 from transformers import (
@@ -581,7 +582,7 @@ def main():
         pad_to_multiple_of=8 if training_args.fp16 else None,
     )
 
-    metric = load_metric("squad_v2" if data_args.version_2_with_negative else "squad")
+    metric = evaluate.load("squad_v2" if data_args.version_2_with_negative else "squad")
 
     def compute_metrics(p: EvalPrediction):
         return metric.compute(predictions=p.predictions, references=p.label_ids)