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

* Alias

* Note alias

* Tests and src

* Rest

* Clean

* Change typing?

* Fix tests

* Deprecation versions

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 \
     "$@"