From 1f843991716bba3b73071f1a2d819b0aebc05375 Mon Sep 17 00:00:00 2001
From: atturaioe <76523524+atturaioe@users.noreply.github.com>
Date: Mon, 1 Aug 2022 14:40:25 +0300
Subject: [PATCH] Migrate metric to Evaluate in Pytorch examples (#18369)

* Migrate metric to Evaluate in pytorch examples

* Remove unused imports
---
 examples/pytorch/_tests_requirements.txt                   | 1 +
 .../audio-classification/run_audio_classification.py       | 3 ++-
 .../image-classification/run_image_classification.py       | 4 ++--
 .../run_image_classification_no_trainer.py                 | 5 +++--
 examples/pytorch/language-modeling/run_clm.py              | 5 +++--
 examples/pytorch/language-modeling/run_mlm.py              | 5 +++--
 examples/pytorch/multiple-choice/run_swag_no_trainer.py    | 5 +++--
 examples/pytorch/question-answering/run_qa.py              | 5 +++--
 examples/pytorch/question-answering/run_qa_beam_search.py  | 5 +++--
 .../question-answering/run_qa_beam_search_no_trainer.py    | 5 +++--
 examples/pytorch/question-answering/run_qa_no_trainer.py   | 5 +++--
 examples/pytorch/question-answering/run_seq2seq_qa.py      | 5 +++--
 .../semantic-segmentation/run_semantic_segmentation.py     | 4 ++--
 .../run_semantic_segmentation_no_trainer.py                | 5 +++--
 .../speech-recognition/run_speech_recognition_ctc.py       | 5 +++--
 .../speech-recognition/run_speech_recognition_seq2seq.py   | 5 +++--
 examples/pytorch/summarization/run_summarization.py        | 5 +++--
 .../pytorch/summarization/run_summarization_no_trainer.py  | 5 +++--
 examples/pytorch/text-classification/run_glue.py           | 7 ++++---
 .../pytorch/text-classification/run_glue_no_trainer.py     | 7 ++++---
 examples/pytorch/text-classification/run_xnli.py           | 5 +++--
 examples/pytorch/token-classification/run_ner.py           | 5 +++--
 .../pytorch/token-classification/run_ner_no_trainer.py     | 5 +++--
 examples/pytorch/translation/run_translation.py            | 5 +++--
 examples/pytorch/translation/run_translation_no_trainer.py | 5 +++--
 25 files changed, 72 insertions(+), 49 deletions(-)

diff --git a/examples/pytorch/_tests_requirements.txt b/examples/pytorch/_tests_requirements.txt
index 6e2dd71816..edaae597ab 100644
--- a/examples/pytorch/_tests_requirements.txt
+++ b/examples/pytorch/_tests_requirements.txt
@@ -23,3 +23,4 @@ torchvision
 jiwer
 librosa
 torch < 1.12
+evaluate
diff --git a/examples/pytorch/audio-classification/run_audio_classification.py b/examples/pytorch/audio-classification/run_audio_classification.py
index 8c380033ba..6c2a6cb880 100644
--- a/examples/pytorch/audio-classification/run_audio_classification.py
+++ b/examples/pytorch/audio-classification/run_audio_classification.py
@@ -26,6 +26,7 @@ import datasets
 import numpy as np
 from datasets import DatasetDict, load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -315,7 +316,7 @@ def main():
         id2label[str(i)] = label
 
     # Load the accuracy metric from the datasets package
-    metric = datasets.load_metric("accuracy")
+    metric = evaluate.load("accuracy")
 
     # Define our compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with
     # `predictions` and `label_ids` fields) and has to return a dictionary string to float.
diff --git a/examples/pytorch/image-classification/run_image_classification.py b/examples/pytorch/image-classification/run_image_classification.py
index 1c75b53b47..f8c2c95f59 100644
--- a/examples/pytorch/image-classification/run_image_classification.py
+++ b/examples/pytorch/image-classification/run_image_classification.py
@@ -19,7 +19,6 @@ import sys
 from dataclasses import dataclass, field
 from typing import Optional
 
-import datasets
 import numpy as np
 import torch
 from datasets import load_dataset
@@ -34,6 +33,7 @@ from torchvision.transforms import (
     ToTensor,
 )
 
+import evaluate
 import transformers
 from transformers import (
     MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING,
@@ -252,7 +252,7 @@ def main():
         id2label[str(i)] = label
 
     # Load the accuracy metric from the datasets package
-    metric = datasets.load_metric("accuracy")
+    metric = evaluate.load("accuracy")
 
     # Define our 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.
diff --git a/examples/pytorch/image-classification/run_image_classification_no_trainer.py b/examples/pytorch/image-classification/run_image_classification_no_trainer.py
index e59cc490f7..773e0265d3 100644
--- a/examples/pytorch/image-classification/run_image_classification_no_trainer.py
+++ b/examples/pytorch/image-classification/run_image_classification_no_trainer.py
@@ -22,7 +22,7 @@ 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 torchvision.transforms import (
     CenterCrop,
@@ -35,6 +35,7 @@ from torchvision.transforms import (
 )
 from tqdm.auto import tqdm
 
+import evaluate
 import transformers
 from accelerate import Accelerator
 from accelerate.logging import get_logger
@@ -415,7 +416,7 @@ def main():
             accelerator.init_trackers("image_classification_no_trainer", experiment_config)
 
     # Get the metric function
-    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
diff --git a/examples/pytorch/language-modeling/run_clm.py b/examples/pytorch/language-modeling/run_clm.py
index 759dfe204a..53052d7671 100755
--- a/examples/pytorch/language-modeling/run_clm.py
+++ b/examples/pytorch/language-modeling/run_clm.py
@@ -30,8 +30,9 @@ from itertools import chain
 from typing import Optional
 
 import datasets
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     CONFIG_MAPPING,
@@ -492,7 +493,7 @@ def main():
                 logits = logits[0]
             return logits.argmax(dim=-1)
 
-        metric = load_metric("accuracy")
+        metric = evaluate.load("accuracy")
 
         def compute_metrics(eval_preds):
             preds, labels = eval_preds
diff --git a/examples/pytorch/language-modeling/run_mlm.py b/examples/pytorch/language-modeling/run_mlm.py
index 29742eba98..dcc8bcd3cd 100755
--- a/examples/pytorch/language-modeling/run_mlm.py
+++ b/examples/pytorch/language-modeling/run_mlm.py
@@ -30,8 +30,9 @@ from itertools import chain
 from typing import Optional
 
 import datasets
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     CONFIG_MAPPING,
@@ -515,7 +516,7 @@ def main():
                 logits = logits[0]
             return logits.argmax(dim=-1)
 
-        metric = load_metric("accuracy")
+        metric = evaluate.load("accuracy")
 
         def compute_metrics(eval_preds):
             preds, labels = eval_preds
diff --git a/examples/pytorch/multiple-choice/run_swag_no_trainer.py b/examples/pytorch/multiple-choice/run_swag_no_trainer.py
index fb22d54b20..d72be5d95b 100755
--- a/examples/pytorch/multiple-choice/run_swag_no_trainer.py
+++ b/examples/pytorch/multiple-choice/run_swag_no_trainer.py
@@ -31,10 +31,11 @@ from typing import Optional, Union
 
 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
@@ -514,7 +515,7 @@ def main():
             accelerator.init_trackers("swag_no_trainer", experiment_config)
 
     # Metrics
-    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
diff --git a/examples/pytorch/question-answering/run_qa.py b/examples/pytorch/question-answering/run_qa.py
index 68fb457616..cddcb4891b 100755
--- a/examples/pytorch/question-answering/run_qa.py
+++ b/examples/pytorch/question-answering/run_qa.py
@@ -25,8 +25,9 @@ from dataclasses import dataclass, field
 from typing import Optional
 
 import datasets
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from trainer_qa import QuestionAnsweringTrainer
 from transformers import (
@@ -593,7 +594,7 @@ def main():
         references = [{"id": ex["id"], "answers": ex[answer_column_name]} for ex in examples]
         return EvalPrediction(predictions=formatted_predictions, label_ids=references)
 
-    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)
diff --git a/examples/pytorch/question-answering/run_qa_beam_search.py b/examples/pytorch/question-answering/run_qa_beam_search.py
index e9db4632f9..1c389e43f3 100755
--- a/examples/pytorch/question-answering/run_qa_beam_search.py
+++ b/examples/pytorch/question-answering/run_qa_beam_search.py
@@ -25,8 +25,9 @@ from dataclasses import dataclass, field
 from typing import Optional
 
 import datasets
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from trainer_qa import QuestionAnsweringTrainer
 from transformers import (
@@ -625,7 +626,7 @@ def main():
         references = [{"id": ex["id"], "answers": ex[answer_column_name]} for ex in examples]
         return EvalPrediction(predictions=formatted_predictions, label_ids=references)
 
-    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)
diff --git a/examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py b/examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py
index eca4c60a6d..e6c66e379a 100644
--- a/examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py
+++ b/examples/pytorch/question-answering/run_qa_beam_search_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
@@ -680,7 +681,7 @@ def main():
         references = [{"id": ex["id"], "answers": ex[answer_column_name]} for ex in examples]
         return EvalPrediction(predictions=formatted_predictions, label_ids=references)
 
-    metric = load_metric("squad_v2" if args.version_2_with_negative else "squad")
+    metric = evaluate.load("squad_v2" if args.version_2_with_negative else "squad")
 
     def create_and_fill_np_array(start_or_end_logits, dataset, max_len):
         """
diff --git a/examples/pytorch/question-answering/run_qa_no_trainer.py b/examples/pytorch/question-answering/run_qa_no_trainer.py
index fe839f28b1..ec86d95b5e 100755
--- a/examples/pytorch/question-answering/run_qa_no_trainer.py
+++ b/examples/pytorch/question-answering/run_qa_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
@@ -696,7 +697,7 @@ def main():
         references = [{"id": ex["id"], "answers": ex[answer_column_name]} for ex in examples]
         return EvalPrediction(predictions=formatted_predictions, label_ids=references)
 
-    metric = load_metric("squad_v2" if args.version_2_with_negative else "squad")
+    metric = evaluate.load("squad_v2" if args.version_2_with_negative else "squad")
 
     # Create and fill numpy array of size len_of_validation_data * max_length_of_output_tensor
     def create_and_fill_np_array(start_or_end_logits, dataset, max_len):
diff --git a/examples/pytorch/question-answering/run_seq2seq_qa.py b/examples/pytorch/question-answering/run_seq2seq_qa.py
index b460f8d9d5..c3c85b31da 100644
--- a/examples/pytorch/question-answering/run_seq2seq_qa.py
+++ b/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)
diff --git a/examples/pytorch/semantic-segmentation/run_semantic_segmentation.py b/examples/pytorch/semantic-segmentation/run_semantic_segmentation.py
index a3fe5734b0..92d07f8f91 100644
--- a/examples/pytorch/semantic-segmentation/run_semantic_segmentation.py
+++ b/examples/pytorch/semantic-segmentation/run_semantic_segmentation.py
@@ -21,7 +21,6 @@ import sys
 from dataclasses import dataclass, field
 from typing import Optional
 
-import datasets
 import numpy as np
 import torch
 from datasets import load_dataset
@@ -30,6 +29,7 @@ from torch import nn
 from torchvision import transforms
 from torchvision.transforms import functional
 
+import evaluate
 import transformers
 from huggingface_hub import hf_hub_download
 from transformers import (
@@ -337,7 +337,7 @@ def main():
     label2id = {v: str(k) for k, v in id2label.items()}
 
     # Load the mean IoU metric from the datasets package
-    metric = datasets.load_metric("mean_iou")
+    metric = evaluate.load("mean_iou")
 
     # Define our 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.
diff --git a/examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py b/examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py
index 35d069a747..c7ac9da3e9 100644
--- a/examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py
+++ b/examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py
@@ -24,13 +24,14 @@ 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 PIL import Image
 from torch.utils.data import DataLoader
 from torchvision import transforms
 from torchvision.transforms import functional
 from tqdm.auto import tqdm
 
+import evaluate
 import transformers
 from accelerate import Accelerator
 from accelerate.logging import get_logger
@@ -500,7 +501,7 @@ def main():
     args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
 
     # Instantiate metric
-    metric = load_metric("mean_iou")
+    metric = evaluate.load("mean_iou")
 
     # We need to initialize the trackers we use, and also store our configuration.
     # We initialize the trackers only on main process because `accelerator.log`
diff --git a/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py b/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py
index 720dfd1ea0..fdbed2b3ab 100755
--- a/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py
+++ b/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py
@@ -28,8 +28,9 @@ from typing import Dict, List, Optional, Union
 import datasets
 import numpy as np
 import torch
-from datasets import DatasetDict, load_dataset, load_metric
+from datasets import DatasetDict, load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -643,7 +644,7 @@ def main():
     # instantiate a data collator and the trainer
 
     # Define evaluation metrics during training, *i.e.* word error rate, character error rate
-    eval_metrics = {metric: load_metric(metric) for metric in data_args.eval_metrics}
+    eval_metrics = {metric: evaluate.load(metric) for metric in data_args.eval_metrics}
 
     # for large datasets it is advised to run the preprocessing on a
     # single machine first with ``args.preprocessing_only`` since there will mostly likely
diff --git a/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py b/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py
index 608bb953c9..0ce8ff0550 100755
--- a/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py
+++ b/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py
@@ -27,8 +27,9 @@ from typing import Any, Dict, List, Optional, Union
 
 import datasets
 import torch
-from datasets import DatasetDict, load_dataset, load_metric
+from datasets import DatasetDict, load_dataset
 
+import evaluate
 import transformers
 from transformers import (
     AutoConfig,
@@ -425,7 +426,7 @@ def main():
         return
 
     # 8. Load Metric
-    metric = load_metric("wer")
+    metric = evaluate.load("wer")
 
     def compute_metrics(pred):
         pred_ids = pred.predictions
diff --git a/examples/pytorch/summarization/run_summarization.py b/examples/pytorch/summarization/run_summarization.py
index 031ac25fa2..636b7a60c4 100755
--- a/examples/pytorch/summarization/run_summarization.py
+++ b/examples/pytorch/summarization/run_summarization.py
@@ -27,8 +27,9 @@ from typing import Optional
 import datasets
 import nltk  # Here to have a nice missing dependency error message early on
 import numpy as np
-from datasets import load_dataset, load_metric
+from datasets import load_dataset
 
+import evaluate
 import transformers
 from filelock import FileLock
 from transformers import (
@@ -598,7 +599,7 @@ def main():
     )
 
     # Metric
-    metric = load_metric("rouge")
+    metric = evaluate.load("rouge")
 
     def postprocess_text(preds, labels):
         preds = [pred.strip() for pred in preds]
diff --git a/examples/pytorch/summarization/run_summarization_no_trainer.py b/examples/pytorch/summarization/run_summarization_no_trainer.py
index 16f24cbdab..dd15b331c9 100644
--- a/examples/pytorch/summarization/run_summarization_no_trainer.py
+++ b/examples/pytorch/summarization/run_summarization_no_trainer.py
@@ -30,10 +30,11 @@ import datasets
 import nltk
 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
@@ -583,7 +584,7 @@ def main():
             accelerator.init_trackers("summarization_no_trainer", experiment_config)
 
     # Metric
-    metric = load_metric("rouge")
+    metric = evaluate.load("rouge")
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
diff --git a/examples/pytorch/text-classification/run_glue.py b/examples/pytorch/text-classification/run_glue.py
index ebd515d24e..556e1f3bbe 100755
--- a/examples/pytorch/text-classification/run_glue.py
+++ b/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.
diff --git a/examples/pytorch/text-classification/run_glue_no_trainer.py b/examples/pytorch/text-classification/run_glue_no_trainer.py
index 4a022db69c..326515c2ec 100644
--- a/examples/pytorch/text-classification/run_glue_no_trainer.py
+++ b/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
diff --git a/examples/pytorch/text-classification/run_xnli.py b/examples/pytorch/text-classification/run_xnli.py
index fc7e1a80ff..2450d24e3d 100755
--- a/examples/pytorch/text-classification/run_xnli.py
+++ b/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.
diff --git a/examples/pytorch/token-classification/run_ner.py b/examples/pytorch/token-classification/run_ner.py
index c2c4274259..13993e58a4 100755
--- a/examples/pytorch/token-classification/run_ner.py
+++ b/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
diff --git a/examples/pytorch/token-classification/run_ner_no_trainer.py b/examples/pytorch/token-classification/run_ner_no_trainer.py
index e6fbc8bfd0..5027f93823 100755
--- a/examples/pytorch/token-classification/run_ner_no_trainer.py
+++ b/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
diff --git a/examples/pytorch/translation/run_translation.py b/examples/pytorch/translation/run_translation.py
index 0f21eb2733..4d03cbf49c 100755
--- a/examples/pytorch/translation/run_translation.py
+++ b/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]
diff --git a/examples/pytorch/translation/run_translation_no_trainer.py b/examples/pytorch/translation/run_translation_no_trainer.py
index 6db6e11c50..914181511b 100644
--- a/examples/pytorch/translation/run_translation_no_trainer.py
+++ b/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]