🚨🚨🚨Deprecate `evaluation_strategy` to `eval_strategy`🚨🚨🚨 (#30190)

* Alias

* Note alias

* Tests and src

* Rest

* Clean

* Change typing?

* Fix tests

* Deprecation versions

docs/source/de/training.md

@@ -128,12 +128,12 @@ Rufen Sie [`~evaluate.compute`] auf `metric` auf, um die Genauigkeit Ihrer Vorhe
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-Wenn Sie Ihre Bewertungsmetriken während der Feinabstimmung überwachen möchten, geben Sie den Parameter `evaluation_strategy` in Ihren Trainingsargumenten an, um die Bewertungsmetrik am Ende jeder Epoche zu ermitteln:
+Wenn Sie Ihre Bewertungsmetriken während der Feinabstimmung überwachen möchten, geben Sie den Parameter `eval_strategy` in Ihren Trainingsargumenten an, um die Bewertungsmetrik am Ende jeder Epoche zu ermitteln:
 
 ```py
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/en/model_memory_anatomy.md

@@ -145,7 +145,7 @@ arguments:
 ```py
 default_args = {
     "output_dir": "tmp",
-    "evaluation_strategy": "steps",
+    "eval_strategy": "steps",
     "num_train_epochs": 1,
     "log_level": "error",
     "report_to": "none",

docs/source/en/tasks/asr.md

@@ -270,7 +270,7 @@ At this point, only three steps remain:
 ...     gradient_checkpointing=True,
 ...     fp16=True,
 ...     group_by_length=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=8,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/en/tasks/audio_classification.md

@@ -221,7 +221,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_mind_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=3e-5,
 ...     per_device_train_batch_size=32,

docs/source/en/tasks/document_question_answering.md

@@ -399,7 +399,7 @@ In this case the `output_dir` will also be the name of the repo where your model
 ...     num_train_epochs=20,
 ...     save_steps=200,
 ...     logging_steps=50,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     learning_rate=5e-5,
 ...     save_total_limit=2,
 ...     remove_unused_columns=False,

docs/source/en/tasks/image_captioning.md

@@ -196,7 +196,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=32,
     gradient_accumulation_steps=2,
     save_total_limit=3,
-    evaluation_strategy="steps",
+    eval_strategy="steps",
     eval_steps=50,
     save_strategy="steps",
     save_steps=50,

docs/source/en/tasks/image_classification.md

@@ -302,7 +302,7 @@ At this point, only three steps remain:
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_food_model",
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=16,

docs/source/en/tasks/knowledge_distillation_for_image_classification.md

@@ -112,7 +112,7 @@ training_args = TrainingArguments(
     fp16=True,
     logging_dir=f"{repo_name}/logs",
     logging_strategy="epoch",
-    evaluation_strategy="epoch",
+    eval_strategy="epoch",
     save_strategy="epoch",
     load_best_model_at_end=True,
     metric_for_best_model="accuracy",

docs/source/en/tasks/language_modeling.md

@@ -249,7 +249,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_clm-model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     weight_decay=0.01,
 ...     push_to_hub=True,

docs/source/en/tasks/masked_language_modeling.md

@@ -238,7 +238,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_mlm_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     num_train_epochs=3,
 ...     weight_decay=0.01,

docs/source/en/tasks/multiple_choice.md

@@ -265,7 +265,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_swag_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     learning_rate=5e-5,

docs/source/en/tasks/question_answering.md

@@ -218,7 +218,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_qa_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/en/tasks/semantic_segmentation.md

@@ -535,7 +535,7 @@ At this point, only three steps remain:
 ...     per_device_train_batch_size=2,
 ...     per_device_eval_batch_size=2,
 ...     save_total_limit=3,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     save_strategy="steps",
 ...     save_steps=20,
 ...     eval_steps=20,

docs/source/en/tasks/sequence_classification.md

@@ -187,7 +187,7 @@ At this point, only three steps remain:
 ...     per_device_eval_batch_size=16,
 ...     num_train_epochs=2,
 ...     weight_decay=0.01,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     push_to_hub=True,

docs/source/en/tasks/summarization.md

@@ -202,7 +202,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_billsum_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/en/tasks/text-to-speech.md

@@ -477,7 +477,7 @@ only look at the loss:
 ...     max_steps=4000,
 ...     gradient_checkpointing=True,
 ...     fp16=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=2,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/en/tasks/token_classification.md

@@ -290,7 +290,7 @@ At this point, only three steps remain:
 ...     per_device_eval_batch_size=16,
 ...     num_train_epochs=2,
 ...     weight_decay=0.01,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     push_to_hub=True,

docs/source/en/tasks/translation.md

@@ -209,7 +209,7 @@ At this point, only three steps remain:
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_opus_books_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/en/tasks/video_classification.md

@@ -354,7 +354,7 @@ Most of the training arguments are self-explanatory, but one that is quite impor
 >>> args = TrainingArguments(
 ...     new_model_name,
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=batch_size,

docs/source/en/trainer.md

@@ -62,7 +62,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=16,
     num_train_epochs=2,
     weight_decay=0.01,
-    evaluation_strategy="epoch",
+    eval_strategy="epoch",
     save_strategy="epoch",
     load_best_model_at_end=True,
     push_to_hub=True,

docs/source/en/training.md

@@ -128,12 +128,12 @@ Call [`~evaluate.compute`] on `metric` to calculate the accuracy of your predict
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-If you'd like to monitor your evaluation metrics during fine-tuning, specify the `evaluation_strategy` parameter in your training arguments to report the evaluation metric at the end of each epoch:
+If you'd like to monitor your evaluation metrics during fine-tuning, specify the `eval_strategy` parameter in your training arguments to report the evaluation metric at the end of each epoch:
 
 ```py
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/es/tasks/asr.md

@@ -260,7 +260,7 @@ En este punto, solo quedan tres pasos:
 ...     gradient_checkpointing=True,
 ...     fp16=True,
 ...     group_by_length=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=8,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/es/tasks/image_captioning.md

@@ -188,7 +188,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=32,
     gradient_accumulation_steps=2,
     save_total_limit=3,
-    evaluation_strategy="steps",
+    eval_strategy="steps",
     eval_steps=50,
     save_strategy="steps",
     save_steps=50,

docs/source/es/tasks/image_classification.md

@@ -143,7 +143,7 @@ Al llegar a este punto, solo quedan tres pasos:
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
 ...     per_device_train_batch_size=16,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     num_train_epochs=4,
 ...     fp16=True,
 ...     save_steps=100,

docs/source/es/tasks/language_modeling.md

@@ -232,7 +232,7 @@ A este punto, solo faltan tres pasos:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     weight_decay=0.01,
 ... )
@@ -338,7 +338,7 @@ A este punto, solo faltan tres pasos:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     num_train_epochs=3,
 ...     weight_decay=0.01,

docs/source/es/tasks/multiple_choice.md

@@ -212,7 +212,7 @@ En este punto, solo quedan tres pasos:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/es/tasks/question_answering.md

@@ -182,7 +182,7 @@ En este punto, solo quedan tres pasos:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/es/tasks/summarization.md

@@ -140,7 +140,7 @@ En este punto, solo faltan tres pasos:
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/es/trainer.md

@@ -60,7 +60,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=16,
     num_train_epochs=2,
     weight_decay=0.01,
-    evaluation_strategy="epoch",
+    eval_strategy="epoch",
     save_strategy="epoch",
     load_best_model_at_end=True,
     push_to_hub=True,

docs/source/es/training.md

@@ -120,12 +120,12 @@ Define la función `compute` en `metric` para calcular el accuracy de tus predic
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-Si quieres controlar tus métricas de evaluación durante el fine-tuning, especifica el parámetro `evaluation_strategy` en tus argumentos de entrenamiento para que el modelo tenga en cuenta la métrica de evaluación al final de cada época:
+Si quieres controlar tus métricas de evaluación durante el fine-tuning, especifica el parámetro `eval_strategy` en tus argumentos de entrenamiento para que el modelo tenga en cuenta la métrica de evaluación al final de cada época:
 
 ```py
 >>> from transformers import TrainingArguments
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/it/migration.md

@@ -167,7 +167,7 @@ Per quanto riguarda la classe `Trainer`:
 - Il metodo `is_world_master` di `Trainer` è deprecato a favore di `is_world_process_zero`.
 
 Per quanto riguarda la classe `TrainingArguments`:
-- L'argomento `evaluate_during_training` di `TrainingArguments` è deprecato a favore di `evaluation_strategy`.
+- L'argomento `evaluate_during_training` di `TrainingArguments` è deprecato a favore di `eval_strategy`.
 
 Per quanto riguarda il modello Transfo-XL:
 - L'attributo di configurazione `tie_weight` di Transfo-XL diventa `tie_words_embeddings`.

docs/source/it/training.md

@@ -121,12 +121,12 @@ Richiama `compute` su `metric` per calcolare l'accuratezza delle tue previsioni.
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-Se preferisci monitorare le tue metriche di valutazione durante il fine-tuning, specifica il parametro `evaluation_strategy` nei tuoi training arguments per restituire le metriche di valutazione ad ogni epoca di addestramento:
+Se preferisci monitorare le tue metriche di valutazione durante il fine-tuning, specifica il parametro `eval_strategy` nei tuoi training arguments per restituire le metriche di valutazione ad ogni epoca di addestramento:
 
 ```py
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/ja/model_memory_anatomy.md

@@ -136,7 +136,7 @@ Tue Jan 11 08:58:05 2022
 ```py
 default_args = {
     "output_dir": "tmp",
-    "evaluation_strategy": "steps",
+    "eval_strategy": "steps",
     "num_train_epochs": 1,
     "log_level": "error",
     "report_to": "none",

docs/source/ja/tasks/asr.md

@@ -270,7 +270,7 @@ MInDS-14 データセットのサンプリング レートは 8000kHz です (
 ...     gradient_checkpointing=True,
 ...     fp16=True,
 ...     group_by_length=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=8,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/ja/tasks/audio_classification.md

@@ -221,7 +221,7 @@ MInDS-14 データセットのサンプリング レートは 8000khz です (
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_mind_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=3e-5,
 ...     per_device_train_batch_size=32,

docs/source/ja/tasks/document_question_answering.md

@@ -403,7 +403,7 @@ end_index 18
 ...     num_train_epochs=20,
 ...     save_steps=200,
 ...     logging_steps=50,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     learning_rate=5e-5,
 ...     save_total_limit=2,
 ...     remove_unused_columns=False,

docs/source/ja/tasks/image_captioning.md

@@ -194,7 +194,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=32,
     gradient_accumulation_steps=2,
     save_total_limit=3,
-    evaluation_strategy="steps",
+    eval_strategy="steps",
     eval_steps=50,
     save_strategy="steps",
     save_steps=50,

docs/source/ja/tasks/image_classification.md

@@ -308,7 +308,7 @@ food["test"].set_transform(preprocess_val)
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_food_model",
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=16,

docs/source/ja/tasks/knowledge_distillation_for_image_classification.md

@@ -112,7 +112,7 @@ training_args = TrainingArguments(
     fp16=True,
     logging_dir=f"{repo_name}/logs",
     logging_strategy="epoch",
-    evaluation_strategy="epoch",
+    eval_strategy="epoch",
     save_strategy="epoch",
     load_best_model_at_end=True,
     metric_for_best_model="accuracy",

docs/source/ja/tasks/language_modeling.md

@@ -246,7 +246,7 @@ Apply the `group_texts` function over the entire dataset:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_clm-model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     weight_decay=0.01,
 ...     push_to_hub=True,

docs/source/ja/tasks/masked_language_modeling.md

@@ -231,7 +231,7 @@ pip install transformers datasets evaluate
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_mlm_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     num_train_epochs=3,
 ...     weight_decay=0.01,

docs/source/ja/tasks/multiple_choice.md

@@ -266,7 +266,7 @@ tokenized_swag = swag.map(preprocess_function, batched=True)
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_swag_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     learning_rate=5e-5,

docs/source/ja/tasks/question_answering.md

@@ -220,7 +220,7 @@ pip install transformers datasets evaluate
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_qa_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ja/tasks/semantic_segmentation.md

@@ -323,7 +323,7 @@ pip install -q datasets transformers evaluate
 ...     per_device_train_batch_size=2,
 ...     per_device_eval_batch_size=2,
 ...     save_total_limit=3,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     save_strategy="steps",
 ...     save_steps=20,
 ...     eval_steps=20,

docs/source/ja/tasks/sequence_classification.md

@@ -324,7 +324,7 @@ pip install -q datasets transformers evaluate
 ...     per_device_train_batch_size=2,
 ...     per_device_eval_batch_size=2,
 ...     save_total_limit=3,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     save_strategy="steps",
 ...     save_steps=20,
 ...     eval_steps=20,

docs/source/ja/tasks/summarization.md

@@ -204,7 +204,7 @@ pip install transformers datasets evaluate rouge_score
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_billsum_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ja/tasks/text-to-speech.md

@@ -477,7 +477,7 @@ SpeechT5 では、モデルのデコーダ部分への入力が 2 分の 1 に
 ...     max_steps=4000,
 ...     gradient_checkpointing=True,
 ...     fp16=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=2,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/ja/tasks/token_classification.md

@@ -288,7 +288,7 @@ pip install transformers datasets evaluate seqeval
 ...     per_device_eval_batch_size=16,
 ...     num_train_epochs=2,
 ...     weight_decay=0.01,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     push_to_hub=True,

docs/source/ja/tasks/translation.md

@@ -208,7 +208,7 @@ pip install transformers datasets evaluate sacrebleu
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_opus_books_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ja/tasks/video_classification.md

@@ -360,7 +360,7 @@ You should probably TRAIN this model on a down-stream task to be able to use it
 >>> args = TrainingArguments(
 ...     new_model_name,
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=batch_size,

docs/source/ja/training.md

@@ -135,12 +135,12 @@ BERTモデルの事前学習済みのヘッドは破棄され、ランダムに
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-評価メトリクスをファインチューニング中に監視したい場合、トレーニング引数で `evaluation_strategy` パラメータを指定して、各エポックの終了時に評価メトリクスを報告します：
+評価メトリクスをファインチューニング中に監視したい場合、トレーニング引数で `eval_strategy` パラメータを指定して、各エポックの終了時に評価メトリクスを報告します：
 
 ```python
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/ko/model_memory_anatomy.md

@@ -132,7 +132,7 @@ Tue Jan 11 08:58:05 2022
 ```py
 default_args = {
     "output_dir": "tmp",
-    "evaluation_strategy": "steps",
+    "eval_strategy": "steps",
     "num_train_epochs": 1,
     "log_level": "error",
     "report_to": "none",

docs/source/ko/tasks/asr.md

@@ -274,7 +274,7 @@ MInDS-14 데이터 세트의 샘플링 레이트는 8000kHz이므로([데이터
 ...     gradient_checkpointing=True,
 ...     fp16=True,
 ...     group_by_length=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=8,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/ko/tasks/audio_classification.md

@@ -221,7 +221,7 @@ MinDS-14 데이터 세트의 샘플링 속도는 8000khz이므로(이 정보는
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_mind_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=3e-5,
 ...     per_device_train_batch_size=32,

docs/source/ko/tasks/document_question_answering.md

@@ -385,7 +385,7 @@ end_index 18
 ...     num_train_epochs=20,
 ...     save_steps=200,
 ...     logging_steps=50,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     learning_rate=5e-5,
 ...     save_total_limit=2,
 ...     remove_unused_columns=False,

docs/source/ko/tasks/image_captioning.md

@@ -201,7 +201,7 @@ training_args = TrainingArguments(
     per_device_eval_batch_size=32,
     gradient_accumulation_steps=2,
     save_total_limit=3,
-    evaluation_strategy="steps",
+    eval_strategy="steps",
     eval_steps=50,
     save_strategy="steps",
     save_steps=50,

docs/source/ko/tasks/image_classification.md

@@ -301,7 +301,7 @@ food["test"].set_transform(preprocess_val)
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_food_model",
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=16,

docs/source/ko/tasks/language_modeling.md

@@ -233,7 +233,7 @@ pip install transformers datasets evaluate
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_clm-model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     weight_decay=0.01,
 ...     push_to_hub=True,

docs/source/ko/tasks/masked_language_modeling.md

@@ -236,7 +236,7 @@ Hugging Face 계정에 로그인하여 모델을 업로드하고 커뮤니티와
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_eli5_mlm_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     num_train_epochs=3,
 ...     weight_decay=0.01,

docs/source/ko/tasks/multiple_choice.md

@@ -265,7 +265,7 @@ tokenized_swag = swag.map(preprocess_function, batched=True)
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_swag_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     learning_rate=5e-5,

docs/source/ko/tasks/question_answering.md

@@ -215,7 +215,7 @@ pip install transformers datasets evaluate
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="my_awesome_qa_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ko/tasks/semantic_segmentation.md

@@ -317,7 +317,7 @@ pip install -q datasets transformers evaluate
 ...     per_device_train_batch_size=2,
 ...     per_device_eval_batch_size=2,
 ...     save_total_limit=3,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     save_strategy="steps",
 ...     save_steps=20,
 ...     eval_steps=20,

docs/source/ko/tasks/sequence_classification.md

@@ -185,7 +185,7 @@ tokenized_imdb = imdb.map(preprocess_function, batched=True)
 ...     per_device_eval_batch_size=16,
 ...     num_train_epochs=2,
 ...     weight_decay=0.01,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     push_to_hub=True,

docs/source/ko/tasks/summarization.md

@@ -211,7 +211,7 @@ Hugging Face 계정에 로그인하면 모델을 업로드하고 커뮤니티에
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_billsum_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ko/tasks/token_classification.md

@@ -288,7 +288,7 @@ Hugging Face 계정에 로그인하여 모델을 업로드하고 커뮤니티에
 ...     per_device_eval_batch_size=16,
 ...     num_train_epochs=2,
 ...     weight_decay=0.01,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     load_best_model_at_end=True,
 ...     push_to_hub=True,

docs/source/ko/tasks/translation.md

@@ -209,7 +209,7 @@ pip install transformers datasets evaluate sacrebleu
 ```py
 >>> training_args = Seq2SeqTrainingArguments(
 ...     output_dir="my_awesome_opus_books_model",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/ko/tasks/video_classification.md

@@ -358,7 +358,7 @@ You should probably TRAIN this model on a down-stream task to be able to use it
 >>> args = TrainingArguments(
 ...     new_model_name,
 ...     remove_unused_columns=False,
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     save_strategy="epoch",
 ...     learning_rate=5e-5,
 ...     per_device_train_batch_size=batch_size,

docs/source/ko/training.md

@@ -129,12 +129,12 @@ rendered properly in your Markdown viewer.
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-미세 튜닝 중에 평가 지표를 모니터링하려면 훈련 인수에 `evaluation_strategy` 파라미터를 지정하여 각 에폭이 끝날 때 평가 지표를 확인할 수 있습니다:
+미세 튜닝 중에 평가 지표를 모니터링하려면 훈련 인수에 `eval_strategy` 파라미터를 지정하여 각 에폭이 끝날 때 평가 지표를 확인할 수 있습니다:
 
 ```py
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### 훈련 하기[[trainer]]

docs/source/pt/tasks/token_classification.md

@@ -180,7 +180,7 @@ Nesse ponto, restam apenas três passos:
 ```py
 >>> training_args = TrainingArguments(
 ...     output_dir="./results",
-...     evaluation_strategy="epoch",
+...     eval_strategy="epoch",
 ...     learning_rate=2e-5,
 ...     per_device_train_batch_size=16,
 ...     per_device_eval_batch_size=16,

docs/source/pt/training.md

@@ -146,13 +146,13 @@ todos os modelos de 🤗 Transformers retornam logits).
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-Se quiser controlar as suas métricas de avaliação durante o fine-tuning, especifique o parâmetro `evaluation_strategy`
+Se quiser controlar as suas métricas de avaliação durante o fine-tuning, especifique o parâmetro `eval_strategy`
 nos seus argumentos de treinamento para que o modelo considere a métrica de avaliação ao final de cada época:
 
 ```py
 >>> from transformers import TrainingArguments
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### Trainer

docs/source/zh/tasks/asr.md

@@ -288,7 +288,7 @@ Wav2Vec2 分词器仅训练了大写字符，因此您需要确保文本与分
 ...     gradient_checkpointing=True,
 ...     fp16=True,
 ...     group_by_length=True,
-...     evaluation_strategy="steps",
+...     eval_strategy="steps",
 ...     per_device_eval_batch_size=8,
 ...     save_steps=1000,
 ...     eval_steps=1000,

docs/source/zh/training.md

@@ -125,12 +125,12 @@ rendered properly in your Markdown viewer.
 ...     return metric.compute(predictions=predictions, references=labels)
 ```
 
-如果您希望在微调过程中监视评估指标，请在您的训练参数中指定 `evaluation_strategy` 参数，以在每个`epoch`结束时展示评估指标：
+如果您希望在微调过程中监视评估指标，请在您的训练参数中指定 `eval_strategy` 参数，以在每个`epoch`结束时展示评估指标：
 
 ```py
 >>> from transformers import TrainingArguments, Trainer
 
->>> training_args = TrainingArguments(output_dir="test_trainer", evaluation_strategy="epoch")
+>>> training_args = TrainingArguments(output_dir="test_trainer", eval_strategy="epoch")
 ```
 
 ### 训练器

examples/flax/language-modeling/README.md

@@ -490,7 +490,7 @@ python3 xla_spawn.py --num_cores ${NUM_TPUS} run_mlm.py --output_dir="./runs" \
     --do_train \
     --do_eval \
     --logging_steps="500" \
-    --evaluation_strategy="epoch" \
+    --eval_strategy="epoch" \
     --report_to="tensorboard" \
     --save_strategy="no"
 ```
@@ -538,7 +538,7 @@ python3 -m torch.distributed.launch --nproc_per_node ${NUM_GPUS} run_mlm.py \
     --do_train \
     --do_eval \
     --logging_steps="500" \
-    --evaluation_strategy="steps" \
+    --eval_strategy="steps" \
     --report_to="tensorboard" \
     --save_strategy="no"
 ```

examples/legacy/seq2seq/finetune.sh

@@ -18,7 +18,7 @@ python finetune_trainer.py \
     --learning_rate=3e-5 \
     --fp16 \
     --do_train --do_eval --do_predict \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --predict_with_generate \
     --n_val 1000 \
     "$@"

examples/legacy/seq2seq/finetune_tpu.sh

@@ -20,7 +20,7 @@ python xla_spawn.py --num_cores $TPU_NUM_CORES \
     finetune_trainer.py \
     --learning_rate=3e-5 \
     --do_train --do_eval \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --prediction_loss_only \
     --n_val 1000 \
     "$@"

examples/legacy/seq2seq/finetune_trainer.py

@@ -271,7 +271,7 @@ def main():
             max_source_length=data_args.max_source_length,
             prefix=model.config.prefix or "",
         )
-        if training_args.do_eval or training_args.evaluation_strategy != EvaluationStrategy.NO
+        if training_args.do_eval or training_args.eval_strategy != EvaluationStrategy.NO
         else None
     )
     test_dataset = (

examples/legacy/seq2seq/train_distil_marian_enro.sh

@@ -32,7 +32,7 @@ python finetune_trainer.py \
     --max_source_length $MAX_LEN --max_target_length $MAX_LEN \
     --val_max_target_length $MAX_TGT_LEN --test_max_target_length $MAX_TGT_LEN \
     --do_train --do_eval --do_predict \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --predict_with_generate --logging_first_step \
     --task translation --label_smoothing_factor 0.1 \
     "$@"

examples/legacy/seq2seq/train_distil_marian_enro_tpu.sh

@@ -33,7 +33,7 @@ python xla_spawn.py --num_cores $TPU_NUM_CORES \
     --max_source_length $MAX_LEN --max_target_length $MAX_LEN \
     --val_max_target_length $MAX_TGT_LEN --test_max_target_length $MAX_TGT_LEN \
     --do_train --do_eval \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --prediction_loss_only \
     --task translation --label_smoothing_factor 0.1 \
     "$@"

examples/legacy/seq2seq/train_distilbart_cnn.sh

@@ -34,6 +34,6 @@ python finetune_trainer.py \
     --logging_first_step \
     --max_target_length 56 --val_max_target_length $MAX_TGT_LEN --test_max_target_length $MAX_TGT_LEN\
     --do_train --do_eval --do_predict \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --predict_with_generate --sortish_sampler \
     "$@"

examples/legacy/seq2seq/train_mbart_cc25_enro.sh

@@ -29,7 +29,7 @@ python finetune_trainer.py \
     --num_train_epochs 6 \
     --save_steps 25000 --eval_steps 25000 --logging_steps 1000 \
     --do_train --do_eval --do_predict \
-    --evaluation_strategy steps \
+    --eval_strategy steps \
     --predict_with_generate --logging_first_step \
     --task translation \
     "$@"

examples/pytorch/README.md

@@ -283,7 +283,7 @@ To enable Neptune logging, in your `TrainingArguments`, set the `report_to` argu
 ```python
 training_args = TrainingArguments(
     "quick-training-distilbert-mrpc",
-    evaluation_strategy="steps",
+    eval_strategy="steps",
     eval_steps=20,
     report_to="neptune",
 )

examples/pytorch/audio-classification/README.md

@@ -50,7 +50,7 @@ python run_audio_classification.py \
     --dataloader_num_workers 4 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --metric_for_best_model accuracy \
@@ -92,7 +92,7 @@ python run_audio_classification.py \
     --dataloader_num_workers 8 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --metric_for_best_model accuracy \

examples/pytorch/image-classification/README.md

@@ -52,7 +52,7 @@ python run_image_classification.py \
     --per_device_eval_batch_size 8 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --save_total_limit 3 \

examples/pytorch/image-pretraining/README.md

@@ -56,7 +56,7 @@ Alternatively, one can decide to further pre-train an already pre-trained (or fi
     --per_device_eval_batch_size 8 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --save_total_limit 3 \
@@ -106,7 +106,7 @@ Next, we can run the script by providing the path to this custom configuration (
     --per_device_eval_batch_size 8 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --save_total_limit 3 \
@@ -172,7 +172,7 @@ python run_mae.py \
     --per_device_eval_batch_size 8 \
     --logging_strategy steps \
     --logging_steps 10 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --load_best_model_at_end True \
     --save_total_limit 3 \

examples/pytorch/semantic-segmentation/README.md

@@ -118,7 +118,7 @@ python run_semantic_segmentation.py \
     --per_device_eval_batch_size 8 \
     --logging_strategy steps \
     --logging_steps 100 \
-    --evaluation_strategy epoch \
+    --eval_strategy epoch \
     --save_strategy epoch \
     --seed 1337
 ```

examples/pytorch/speech-recognition/README.md

@@ -76,7 +76,7 @@ python run_speech_recognition_ctc.py \
 	--gradient_accumulation_steps="2" \
 	--learning_rate="3e-4" \
 	--warmup_steps="500" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--length_column_name="input_length" \
 	--save_steps="400" \
@@ -111,7 +111,7 @@ torchrun \
 	--per_device_train_batch_size="4" \
 	--learning_rate="3e-4" \
 	--warmup_steps="500" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--length_column_name="input_length" \
 	--save_steps="400" \
@@ -162,7 +162,7 @@ However, the `--shuffle_buffer_size` argument controls how many examples we can
 	--gradient_accumulation_steps="2" \
 	--learning_rate="5e-4" \
 	--warmup_steps="500" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--save_steps="500" \
 	--eval_steps="500" \
@@ -293,7 +293,7 @@ python run_speech_recognition_ctc.py \
 	--per_device_train_batch_size="32" \
 	--learning_rate="1e-3" \
 	--warmup_steps="100" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--length_column_name="input_length" \
 	--save_steps="200" \
@@ -330,7 +330,7 @@ python run_speech_recognition_ctc.py \
 	--per_device_train_batch_size="32" \
 	--learning_rate="1e-3" \
 	--warmup_steps="100" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--length_column_name="input_length" \
 	--save_steps="200" \
@@ -378,7 +378,7 @@ python run_speech_recognition_seq2seq.py \
 	--logging_steps="25" \
 	--learning_rate="1e-5" \
 	--warmup_steps="500" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--eval_steps="1000" \
 	--save_strategy="steps" \
 	--save_steps="1000" \
@@ -419,7 +419,7 @@ torchrun \
 	--logging_steps="25" \
 	--learning_rate="1e-5" \
 	--warmup_steps="500" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--eval_steps="1000" \
 	--save_strategy="steps" \
 	--save_steps="1000" \
@@ -547,7 +547,7 @@ python run_speech_recognition_seq2seq.py \
 	--gradient_accumulation_steps="8" \
 	--learning_rate="3e-4" \
 	--warmup_steps="400" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="text" \
 	--save_steps="400" \
 	--eval_steps="400" \
@@ -589,7 +589,7 @@ torchrun \
 	--gradient_accumulation_steps="1" \
 	--learning_rate="3e-4" \
 	--warmup_steps="400" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="text" \
 	--save_steps="400" \
 	--eval_steps="400" \

examples/research_projects/codeparrot/examples/train_complexity_predictor.py

@@ -100,7 +100,7 @@ def main():
         output_dir=args.output_dir,
         learning_rate=args.learning_rate,
         lr_scheduler_type=args.lr_scheduler_type,
-        evaluation_strategy="epoch",
+        eval_strategy="epoch",
         save_strategy="epoch",
         logging_strategy="epoch",
         per_device_train_batch_size=args.batch_size,

examples/research_projects/layoutlmv3/README.md

@@ -32,7 +32,7 @@ python run_funsd_cord.py \
   --do_train \
   --do_eval \
   --max_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --eval_steps 100 \
   --learning_rate 1e-5 \
   --load_best_model_at_end \
@@ -57,7 +57,7 @@ python run_funsd_cord.py \
   --do_train \
   --do_eval \
   --max_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --eval_steps 100 \
   --learning_rate 5e-5 \
   --load_best_model_at_end \

examples/research_projects/robust-speech-event/README.md

@@ -362,7 +362,7 @@ echo '''python run_speech_recognition_ctc.py \
 	--per_device_train_batch_size="2" \
 	--learning_rate="3e-4" \
 	--save_total_limit="1" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--length_column_name="input_length" \
 	--save_steps="5" \
@@ -438,7 +438,7 @@ echo '''python run_speech_recognition_ctc.py \
 	--learning_rate="7.5e-5" \
 	--warmup_steps="2000" \
 	--length_column_name="input_length" \
-	--evaluation_strategy="steps" \
+	--eval_strategy="steps" \
 	--text_column_name="sentence" \
 	--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” <20> — ’ … – \
 	--save_steps="500" \

examples/research_projects/self-training-text-classification/README.md

@@ -51,7 +51,7 @@ parameters_dict = {
     'train_file': os.path.join(data_dir, 'train.csv'),
     'infer_file': os.path.join(data_dir, 'infer.csv'),
     'eval_file': os.path.join(data_dir, 'eval.csv'),
-    'evaluation_strategy': 'steps',
+    'eval_strategy': 'steps',
     'task_name': 'scitail',
     'label_list': ['entails', 'neutral'],
     'per_device_train_batch_size': 32,

examples/research_projects/self-training-text-classification/finetuning.py

@@ -190,7 +190,7 @@ class FTTrainingArguments:
             )
         },
     )
-    evaluation_strategy: Optional[str] = dataclasses.field(
+    eval_strategy: Optional[str] = dataclasses.field(
         default="no",
         metadata={
             "help": 'The evaluation strategy to adopt during training. Possible values are: ["no", "step", "epoch]'
@@ -198,7 +198,7 @@ class FTTrainingArguments:
     )
     eval_steps: Optional[int] = dataclasses.field(
         default=1,
-        metadata={"help": 'Number of update steps between two evaluations if `evaluation_strategy="steps"`.'},
+        metadata={"help": 'Number of update steps between two evaluations if `eval_strategy="steps"`.'},
     )
     eval_metric: Optional[str] = dataclasses.field(
         default="accuracy", metadata={"help": "The evaluation metric used for the task."}
@@ -265,7 +265,7 @@ def train(args, accelerator, model, tokenizer, train_dataloader, optimizer, lr_s
                 # Evaluate during training
                 if (
                     eval_dataloader is not None
-                    and args.evaluation_strategy == IntervalStrategy.STEPS.value
+                    and args.eval_strategy == IntervalStrategy.STEPS.value
                     and args.eval_steps > 0
                     and completed_steps % args.eval_steps == 0
                 ):
@@ -331,7 +331,7 @@ def train(args, accelerator, model, tokenizer, train_dataloader, optimizer, lr_s
                 break
 
         # Evaluate during training
-        if eval_dataloader is not None and args.evaluation_strategy == IntervalStrategy.EPOCH.value:
+        if eval_dataloader is not None and args.eval_strategy == IntervalStrategy.EPOCH.value:
             accelerator.wait_for_everyone()
             new_checkpoint = f"checkpoint-{IntervalStrategy.EPOCH.value}-{epoch}"
             new_eval_result = evaluate(args, accelerator, eval_dataloader, "eval", model, new_checkpoint)[
@@ -571,7 +571,7 @@ def finetune(accelerator, model_name_or_path, train_file, output_dir, **kwargs):
     assert args.train_file is not None
     data_files[Split.TRAIN.value] = args.train_file
 
-    if args.do_eval or args.evaluation_strategy != IntervalStrategy.NO.value:
+    if args.do_eval or args.eval_strategy != IntervalStrategy.NO.value:
         assert args.eval_file is not None
         data_files[Split.EVAL.value] = args.eval_file
 

examples/research_projects/self-training-text-classification/run.sh

@@ -60,7 +60,7 @@ parameters_dict = {
   'train_file': os.path.join(data_dir, '${TRAIN_FILE}'),
   'infer_file': os.path.join(data_dir, '${INFER_FILE}'),
   'eval_file': os.path.join(data_dir, '${EVAL_FILE}'),
-  'evaluation_strategy': 'steps',
+  'eval_strategy': 'steps',
   'task_name': 'scitail',
   'label_list': ['entails', 'neutral'],
   'per_device_train_batch_size': 32,

examples/research_projects/self-training-text-classification/selftraining.py

@@ -79,7 +79,7 @@ class STTrainingArguments:
     eval_metric: Optional[str] = dataclasses.field(
         default="accuracy", metadata={"help": "The evaluation metric used for the task."}
     )
-    evaluation_strategy: Optional[str] = dataclasses.field(
+    eval_strategy: Optional[str] = dataclasses.field(
         default="no",
         metadata={
             "help": 'The evaluation strategy to adopt during training. Possible values are: ["no", "step", "epoch]'
@@ -208,7 +208,7 @@ def selftrain(model_name_or_path, train_file, infer_file, output_dir, **kwargs):
     data_files["train"] = args.train_file
     data_files["infer"] = args.infer_file
 
-    if args.evaluation_strategy != IntervalStrategy.NO.value:
+    if args.eval_strategy != IntervalStrategy.NO.value:
         assert args.eval_file is not None
         data_files["eval"] = args.eval_file
 
@@ -267,7 +267,7 @@ def selftrain(model_name_or_path, train_file, infer_file, output_dir, **kwargs):
             "label_list": args.label_list,
             "output_dir": current_output_dir,
             "eval_metric": args.eval_metric,
-            "evaluation_strategy": args.evaluation_strategy,
+            "eval_strategy": args.eval_strategy,
             "early_stopping_patience": args.early_stopping_patience,
             "early_stopping_threshold": args.early_stopping_threshold,
             "seed": args.seed,
@@ -341,7 +341,7 @@ def selftrain(model_name_or_path, train_file, infer_file, output_dir, **kwargs):
 
         data_files["train_pseudo"] = os.path.join(next_data_dir, f"train_pseudo.{args.data_file_extension}")
 
-        if args.evaluation_strategy != IntervalStrategy.NO.value:
+        if args.eval_strategy != IntervalStrategy.NO.value:
             new_eval_result = eval_result
 
             if best_iteration is None:

examples/research_projects/tapex/README.md

@@ -71,7 +71,7 @@ python run_wikisql_with_tapex.py \
   --eval_steps 1000 \
   --save_steps 1000 \
   --warmup_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --predict_with_generate \
   --num_beams 5 \
   --weight_decay 1e-2 \
@@ -101,7 +101,7 @@ python run_wikisql_with_tapex.py \
   --eval_steps 1000 \
   --save_steps 1000 \
   --warmup_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --predict_with_generate \
   --num_beams 5 \
   --weight_decay 1e-2 \
@@ -132,7 +132,7 @@ python run_wikitablequestions_with_tapex.py \
   --eval_steps 1000 \
   --save_steps 1000 \
   --warmup_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --predict_with_generate \
   --num_beams 5 \
   --weight_decay 1e-2 \
@@ -162,7 +162,7 @@ python run_wikitablequestions_with_tapex.py \
   --eval_steps 1000 \
   --save_steps 1000 \
   --warmup_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --predict_with_generate \
   --num_beams 5 \
   --weight_decay 1e-2 \
@@ -223,7 +223,7 @@ python run_tabfact_with_tapex.py \
   --learning_rate 3e-5 \
   --eval_steps 1000 \
   --save_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --weight_decay 1e-2 \
   --max_steps 30000 \
   --max_grad_norm 0.1
@@ -252,7 +252,7 @@ python run_tabfact_with_tapex.py \
   --learning_rate 3e-5 \
   --eval_steps 1000 \
   --save_steps 1000 \
-  --evaluation_strategy steps \
+  --eval_strategy steps \
   --weight_decay 1e-2 \
   --max_steps 30000 \
   --max_grad_norm 0.1

examples/research_projects/wav2vec2/FINE_TUNE_XLSR_WAV2VEC2.md

@@ -182,7 +182,7 @@ Here we will run the script on the *Turkish* Common Voice dataset for demonstrat
 		--per_device_train_batch_size="16" \
 		--learning_rate="3e-4" \
 		--warmup_steps="500" \
-		--evaluation_strategy="steps" \
+		--eval_strategy="steps" \
 		--save_steps="400" \
 		--eval_steps="400" \
 		--logging_steps="400" \
@@ -209,7 +209,7 @@ Here we will run the script on the *Turkish* Common Voice dataset for demonstrat
 		--per_device_train_batch_size="16" \
 		--learning_rate="3e-4" \
 		--warmup_steps="500" \
-		--evaluation_strategy="steps" \
+		--eval_strategy="steps" \
 		--save_steps="400" \
 		--eval_steps="400" \
 		--logging_steps="400" \

examples/research_projects/wav2vec2/README.md

@@ -18,7 +18,7 @@ python run_asr.py \
 --num_train_epochs="30" \
 --per_device_train_batch_size="20" \
 --per_device_eval_batch_size="20" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_steps="500" \
 --eval_steps="100" \
 --logging_steps="50" \
@@ -73,7 +73,7 @@ python run_asr.py \
 --per_device_train_batch_size="1" \
 --per_device_eval_batch_size="1" \
 --gradient_accumulation_steps="8" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_steps="500" \
 --eval_steps="100" \
 --logging_steps="50" \
@@ -152,7 +152,7 @@ ZeRO-2:
 PYTHONPATH=../../../src deepspeed --num_gpus 2 \
 run_asr.py \
 --output_dir=output_dir --num_train_epochs=2 --per_device_train_batch_size=2 \
---per_device_eval_batch_size=2 --evaluation_strategy=steps --save_steps=500 --eval_steps=100 \
+--per_device_eval_batch_size=2 --eval_strategy=steps --save_steps=500 --eval_steps=100 \
 --logging_steps=5 --learning_rate=5e-4 --warmup_steps=3000 \
 --model_name_or_path=patrickvonplaten/wav2vec2_tiny_random_robust \
 --dataset_name=hf-internal-testing/librispeech_asr_dummy --dataset_config_name=clean \
@@ -176,7 +176,7 @@ ZeRO-3:
 PYTHONPATH=../../../src deepspeed --num_gpus 2 \
 run_asr.py \
 --output_dir=output_dir --num_train_epochs=2 --per_device_train_batch_size=2 \
---per_device_eval_batch_size=2 --evaluation_strategy=steps --save_steps=500 --eval_steps=100 \
+--per_device_eval_batch_size=2 --eval_strategy=steps --save_steps=500 --eval_steps=100 \
 --logging_steps=5 --learning_rate=5e-4 --warmup_steps=3000 \
 --model_name_or_path=patrickvonplaten/wav2vec2_tiny_random_robust \
 --dataset_name=hf-internal-testing/librispeech_asr_dummy --dataset_config_name=clean \

examples/research_projects/wav2vec2/finetune_base_100.sh

@@ -4,7 +4,7 @@ python run_asr.py \
 --num_train_epochs="30" \
 --per_device_train_batch_size="32" \
 --per_device_eval_batch_size="32" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_total_limit="3" \
 --save_steps="500" \
 --eval_steps="100" \

examples/research_projects/wav2vec2/finetune_base_timit_asr.sh

@@ -4,7 +4,7 @@ python run_asr.py \
 --num_train_epochs="30" \
 --per_device_train_batch_size="20" \
 --per_device_eval_batch_size="20" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_steps="500" \
 --eval_steps="100" \
 --logging_steps="50" \

examples/research_projects/wav2vec2/finetune_large_lv60_100.sh

@@ -4,7 +4,7 @@ python run_asr.py \
 --num_train_epochs="30" \
 --per_device_train_batch_size="16" \
 --per_device_eval_batch_size="16" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_total_limit="3" \
 --save_steps="500" \
 --eval_steps="100" \

examples/research_projects/wav2vec2/finetune_large_lv60_timit_asr.sh

@@ -5,7 +5,7 @@ python run_asr.py \
 --per_device_train_batch_size="2" \
 --per_device_eval_batch_size="2" \
 --gradient_accumulation_steps="4" \
---evaluation_strategy="steps" \
+--eval_strategy="steps" \
 --save_steps="500" \
 --eval_steps="100" \
 --logging_steps="50" \