Migrate metric to Evaluate in Pytorch examples (#18369)

* Migrate metric to Evaluate in pytorch examples

* Remove unused imports

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.

examples/pytorch/token-classification/run_ner.py

@@ -27,8 +27,9 @@ from typing import Optional
 
 import datasets
 import numpy as np
-from datasets import ClassLabel, load_dataset, load_metric
+from datasets import ClassLabel, load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -504,7 +505,7 @@ def main():
     data_collator = DataCollatorForTokenClassification(tokenizer, pad_to_multiple_of=8 if training_args.fp16 else None)
 
     # Metrics
-    metric = load_metric("seqeval")
+    metric = evaluate.load("seqeval")
 
     def compute_metrics(p):
         predictions, labels = p

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -28,10 +28,11 @@ from pathlib import Path
 
 import datasets
 import torch
-from datasets import ClassLabel, load_dataset, load_metric
+from datasets import ClassLabel, 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
@@ -580,7 +581,7 @@ def main():
             accelerator.init_trackers("ner_no_trainer", experiment_config)
 
     # Metrics
-    metric = load_metric("seqeval")
+    metric = evaluate.load("seqeval")
 
     def get_labels(predictions, references):
         # Transform predictions and references tensos to numpy arrays

examples/pytorch/translation/run_translation.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,
@@ -522,7 +523,7 @@ def main():
         )
 
     # Metric
-    metric = load_metric("sacrebleu")
+    metric = evaluate.load("sacrebleu")
 
     def postprocess_text(preds, labels):
         preds = [pred.strip() for pred in preds]

examples/pytorch/translation/run_translation_no_trainer.py

@@ -29,10 +29,11 @@ from pathlib import Path
 import datasets
 import numpy as np
 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
@@ -562,7 +563,7 @@ def main():
             experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
             accelerator.init_trackers("translation_no_trainer", experiment_config)
 
-    metric = load_metric("sacrebleu")
+    metric = evaluate.load("sacrebleu")
 
     def postprocess_text(preds, labels):
         preds = [pred.strip() for pred in preds]