Migrate metric to Evaluate in Pytorch examples (#18369)

* Migrate metric to Evaluate in pytorch examples

* Remove unused imports

examples/pytorch/text-classification/run_glue.py

@@ -25,8 +25,9 @@ from typing import Optional
 
 import datasets
 import numpy as np
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -480,9 +481,9 @@ def main():
 
     # Get the metric function
     if data_args.task_name is not None:
-        metric = load_metric("glue", data_args.task_name)
+        metric = evaluate.load("glue", data_args.task_name)
     else:
-        metric = load_metric("accuracy")
+        metric = evaluate.load("accuracy")
 
     # You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
     # predictions and label_ids field) and has to return a dictionary string to float.

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -23,10 +23,11 @@ from pathlib import Path
 
 import datasets
 import torch
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 
+import evaluate
 import transformers
 from accelerate import Accelerator
 from accelerate.logging import get_logger
@@ -466,9 +467,9 @@ def main():
 
     # Get the metric function
     if args.task_name is not None:
-        metric = load_metric("glue", args.task_name)
+        metric = evaluate.load("glue", args.task_name)
     else:
-        metric = load_metric("accuracy")
+        metric = evaluate.load("accuracy")
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

examples/pytorch/text-classification/run_xnli.py

@@ -26,8 +26,9 @@ from typing import Optional
 
 import datasets
 import numpy as np
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -349,7 +350,7 @@ def main():
             )
 
     # Get the metric function
-    metric = load_metric("xnli")
+    metric = evaluate.load("xnli")
 
     # You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
     # predictions and label_ids field) and has to return a dictionary string to float.