mirror of
https://github.com/huggingface/transformers.git
synced 2025-10-20 17:13:56 +08:00
deprecate overwrite_output_dir (#41323)
* dep * style * rm * wut * style
This commit is contained in:
Marc Sun
GitHub
@ -153,7 +153,7 @@ You are not required to read the following guidelines before opening an issue. H
|
||||
cd examples/seq2seq
|
||||
torchrun --nproc_per_node=2 ./finetune_trainer.py \
|
||||
--model_name_or_path sshleifer/distill-mbart-en-ro-12-4 --data_dir wmt_en_ro \
|
||||
--output_dir output_dir --overwrite_output_dir \
|
||||
--output_dir output_dir \
|
||||
--do_train --n_train 500 --num_train_epochs 1 \
|
||||
--per_device_train_batch_size 1 --freeze_embeds \
|
||||
--src_lang en_XX --tgt_lang ro_RO --task translation \
|
||||
|
||||
@ -93,7 +93,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -117,7 +116,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -140,7 +138,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -197,7 +194,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -225,7 +221,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -239,8 +234,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
خيار آخر مفيد لتمكينه هو استئناف التدريب من نقطة تفتيش سابقة. سيضمن ذلك أنك تستطيع الاستمرار من حيث توقفت دون البدء من جديد إذا تم مقاطعة تدريبك. هناك طريقتان لاستئناف التدريب من نقطة تفتيش.
|
||||
|
||||
تستخدم الطريقة الأولى المعلمة `output_dir previous_output_dir` لاستئناف التدريب من أحدث نقطة تفتيش مخزنة في `output_dir`. في هذه الحالة، يجب عليك إزالة `overwrite_output_dir`:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -252,24 +245,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
تستخدم الطريقة الثانية معلمة `resume_from_checkpoint path_to_specific_checkpoint` لاستئناف التدريب من مجلد نقطة تفتيش محددة.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -301,6 +276,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -611,7 +611,6 @@ accelerate launch \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
يمكنك أيضًا تحديد المعلمات من ملف `config_file.yaml` مباشرة في سطر الأوامر:
|
||||
@ -634,7 +633,6 @@ accelerate launch --num_processes=2 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
اطلع على برنامج تعليمي [Launching your Accelerate scripts](https://huggingface.co/docs/accelerate/basic_tutorials/launch) لمعرفة المزيد حول `accelerate_launch` والتكوينات المخصصة.
|
||||
|
||||
@ -98,7 +98,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -122,7 +121,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -144,7 +142,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -201,7 +198,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -229,7 +225,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -243,8 +238,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
Eine weitere hilfreiche Option, die Sie aktivieren können, ist die Wiederaufnahme des Trainings von einem früheren Kontrollpunkt aus. Auf diese Weise können Sie im Falle einer Unterbrechung Ihres Trainings dort weitermachen, wo Sie aufgehört haben, ohne von vorne beginnen zu müssen. Es gibt zwei Methoden, um das Training von einem Kontrollpunkt aus wieder aufzunehmen.
|
||||
|
||||
Die erste Methode verwendet das Argument `output_dir previous_output_dir`, um das Training ab dem letzten in `output_dir` gespeicherten Kontrollpunkt wieder aufzunehmen. In diesem Fall sollten Sie `overwrite_output_dir` entfernen:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -256,24 +249,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
Die zweite Methode verwendet das Argument `Resume_from_checkpoint path_to_specific_checkpoint`, um das Training ab einem bestimmten Checkpoint-Ordner wieder aufzunehmen.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -305,6 +280,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -593,7 +593,7 @@ To deploy DeepSpeed on multiple GPUs, add `--num_gpus`. You don't need to add `-
|
||||
deepspeed --num_gpus=2 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero3.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -616,7 +616,7 @@ To deploy DeepSpeed on a single GPU, add `--num_gpus`. You don't need to add `--
|
||||
deepspeed --num_gpus=1 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero2.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
|
||||
@ -61,9 +61,8 @@ The example below fine-tunes [T5-small](https://huggingface.co/google-t5/t5-smal
|
||||
|
||||
The example script downloads and preprocesses a dataset, and then fine-tunes it with [`Trainer`] with a supported model architecture.
|
||||
|
||||
Resuming training from a checkpoint is very useful if training is interrupted because you don't have to start over again. There are two ways to resume training from a checkpoint.
|
||||
Resuming training from a checkpoint is very useful if training is interrupted because you don't have to start over again:
|
||||
|
||||
* `--output dir previous_output_dir` resumes training from the latest checkpoint stored in `output_dir`. Remove the `--overwrite_output_dir` parameter if you're using this method.
|
||||
* `--resume_from_checkpoint path_to_specific_checkpoint` resumes training from a specific checkpoint folder.
|
||||
|
||||
Share your model on the [Hub](https://huggingface.co/) with the `--push_to_hub` parameter. It creates a repository and uploads the model to the folder name specified in `--output_dir`. You could also use the `--push_to_hub_model_id` parameter to specify the repository name.
|
||||
@ -85,9 +84,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--push_to_hub \
|
||||
--push_to_hub_model_id finetuned-t5-cnn_dailymail \
|
||||
# remove if using `output_dir previous_output_dir`
|
||||
# --overwrite_output_dir \
|
||||
--output_dir previous_output_dir \
|
||||
# --resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate \
|
||||
```
|
||||
@ -168,7 +164,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate \
|
||||
|
||||
@ -361,8 +361,7 @@ accelerate launch \
|
||||
--per_device_train_batch_size 16 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
--output_dir /tmp/$TASK_NAME/
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
|
||||
@ -98,7 +98,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -122,7 +121,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -144,7 +142,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -201,7 +198,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -229,7 +225,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -243,8 +238,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
Otra opción útil para habilitar es reanudar el entrenamiento desde un punto de control anterior. Esto asegurará que puedas continuar donde lo dejaste sin comenzar de nuevo si tu entrenamiento se interrumpe. Hay dos métodos para reanudar el entrenamiento desde un punto de control.
|
||||
|
||||
El primer método utiliza el argumento `output_dir previous_output_dir` para reanudar el entrenamiento desde el último punto de control almacenado en `output_dir`. En este caso, debes eliminar `overwrite_output_dir`:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -256,24 +249,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
El segundo método utiliza el argumento `resume_from_checkpoint path_to_specific_checkpoint` para reanudar el entrenamiento desde una carpeta de punto de control específica.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -305,6 +280,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -381,7 +381,6 @@ accelerate launch \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
También puedes especificar los parámetros del archivo config_file.yaml directamente en la línea de comandos:
|
||||
@ -404,7 +403,6 @@ accelerate launch --num_processes=2 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
Consulta el tutorial [Lanzamiento de tus scripts con Accelerate](https://huggingface.co/docs/accelerate/basic_tutorials/launch) para obtener más información sobre `accelerate_launch` y las configuraciones personalizadas.
|
||||
|
||||
@ -100,7 +100,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -124,7 +123,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -147,7 +145,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -204,7 +201,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -231,7 +227,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -243,9 +238,7 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
## Reprendre l'entraînement à partir d'un point de contrôle
|
||||
|
||||
Une autre option utile est de reprendre l'entraînement à partir d'un point de contrôle précédent. Cela vous permettra de reprendre là où vous vous étiez arrêté sans recommencer si votre entraînement est interrompu. Il existe deux méthodes pour reprendre l'entraînement à partir d'un point de contrôle.
|
||||
|
||||
La première méthode utilise l'argument `output_dir previous_output_dir` pour reprendre l'entraînement à partir du dernier point de contrôle stocké dans `output_dir`. Dans ce cas, vous devez supprimer l'argument `overwrite_output_dir`.
|
||||
Une autre option utile est de reprendre l'entraînement à partir d'un point de contrôle précédent. Cela vous permettra de reprendre là où vous vous étiez arrêté sans recommencer si votre entraînement est interrompu:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
@ -258,24 +251,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
La seconde méthode utilise l'argument `resume_from_checkpoint path_to_specific_checkpoint` pour reprendre l'entraînement à partir d'un dossier de point de contrôle spécifique.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -308,6 +283,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -98,7 +98,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -122,7 +121,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -144,7 +142,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -201,7 +198,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -229,7 +225,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -243,8 +238,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
Un'altra utile opzione è riavviare un addestramento da un checkpoint precedente. Questo garantirà che tu possa riprendere da dove hai interrotto senza ricominciare se l'addestramento viene interrotto. Ci sono due metodi per riavviare l'addestramento da un checkpoint:
|
||||
|
||||
Il primo metodo usa l'argomento `output_dir previous_output_dir` per riavviare l'addestramento dall'ultima versione del checkpoint contenuto in `output_dir`. In questo caso, dovresti rimuovere `overwrite_output_dir`:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -256,24 +249,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
Il secondo metodo usa l'argomento `resume_from_checkpoint path_to_specific_checkpoint` per riavviare un addestramento da una specifica cartella di checkpoint.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -305,6 +280,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -188,7 +188,7 @@ deepspeed --num_gpus=2 your_program.py <normal cl args> --deepspeed ds_config.js
|
||||
deepspeed examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero3.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -211,7 +211,7 @@ DeepSpeed 関連の引数が 2 つありますが、簡単にするためであ
|
||||
deepspeed --num_gpus=1 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero2.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -1789,7 +1789,7 @@ deepspeed examples/pytorch/translation/run_translation.py \
|
||||
--model_name_or_path google-t5/t5-small --output_dir output_dir \
|
||||
--do_eval --max_eval_samples 50 --warmup_steps 50 \
|
||||
--max_source_length 128 --val_max_target_length 128 \
|
||||
--overwrite_output_dir --per_device_eval_batch_size 4 \
|
||||
--per_device_eval_batch_size 4 \
|
||||
--predict_with_generate --dataset_config "ro-en" --fp16 \
|
||||
--source_lang en --target_lang ro --dataset_name wmt16 \
|
||||
--source_prefix "translate English to Romanian: "
|
||||
|
||||
@ -534,7 +534,6 @@ python examples/pytorch/text-classification/run_glue.py \
|
||||
--learning_rate 2e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
**注意すべきいくつかの注意事項**
|
||||
@ -669,7 +668,6 @@ accelerate launch \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
4. `accelerate launch`するための cmd 引数を直接使用することもできます。上の例は次のようにマッピングされます。
|
||||
@ -694,7 +692,6 @@ accelerate launch --num_processes=2 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
詳細については、🤗 Accelerate CLI ガイドを参照してください: [🤗 Accelerate スクリプトの起動](https://huggingface.co/docs/accelerate/basic_tutorials/launch)。
|
||||
|
||||
@ -104,7 +104,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -131,7 +130,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -153,7 +151,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -212,7 +209,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -240,7 +236,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -254,8 +249,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
以前のチェックポイントからトレーニングを再開するための役立つオプションもあります。これにより、トレーニングが中断された場合でも、最初からやり直すことなく、中断したところから再開できます。チェックポイントからトレーニングを再開するための2つの方法があります。
|
||||
|
||||
最初の方法は、`output_dir previous_output_dir` 引数を使用して、`output_dir` に保存された最新のチェックポイントからトレーニングを再開する方法です。この場合、`overwrite_output_dir` を削除する必要があります：
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -267,25 +260,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
2番目の方法では、`resume_from_checkpoint path_to_specific_checkpoint` 引数を使用して、特定のチェックポイントフォルダからトレーニングを再開します。
|
||||
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -319,7 +293,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
|
||||
@ -590,7 +590,7 @@ bf16은 설정 파일에서 설정하거나 다음 인수를 전달하면 명령
|
||||
deepspeed --num_gpus=2 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero3.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -605,7 +605,7 @@ deepspeed --num_gpus=2 examples/pytorch/translation/run_translation.py \
|
||||
deepspeed --num_gpus=1 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero2.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
|
||||
@ -54,8 +54,7 @@ python examples/pytorch/text-classification/run_glue.py \
|
||||
--per_device_train_batch_size 32 \
|
||||
--learning_rate 2e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
--output_dir /tmp/$TASK_NAME/
|
||||
```
|
||||
|
||||
`gloco`와 `nccl`과 같은 [분산 학습 백엔드](https://pytorch.org/docs/stable/distributed.html#backends)는 `mps` 장치에서 지원되지 않으므로, MPS 백엔드에서는 단일 GPU로만 학습이 가능합니다.
|
||||
|
||||
@ -106,7 +106,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -131,7 +130,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -156,7 +154,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -216,7 +213,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -245,7 +241,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -262,8 +257,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
이렇게 하면 훈련이 중단되더라도 처음부터 다시 시작하지 않고 중단한 부분부터 다시 시작할 수 있습니다.
|
||||
체크포인트에서 훈련을 재개하는 방법에는 두 가지가 있습니다.
|
||||
|
||||
첫 번째는 `output_dir previous_output_dir` 인수를 사용하여 `output_dir`에 저장된 최신 체크포인트부터 훈련을 재개하는 방법입니다.
|
||||
이 경우 `overwrite_output_dir`을 제거해야 합니다:
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -275,24 +268,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
두 번째는 `resume_from_checkpoint path_to_specific_checkpoint` 인수를 사용하여 특정 체크포인트 폴더에서 훈련을 재개하는 방법입니다.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -325,6 +300,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -505,7 +505,6 @@ accelerate launch \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
`config_file.yaml` 파일의 매개변수를 직접 지정할 수도 있습니다:
|
||||
@ -528,7 +527,6 @@ accelerate launch --num_processes=2 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
`accelerate_launch`와 사용자 정의 구성에 대해 더 알아보려면 [Accelerate 스크립트 실행](https://huggingface.co/docs/accelerate/basic_tutorials/launch) 튜토리얼을 확인하세요.
|
||||
@ -99,7 +99,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -123,7 +122,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -145,7 +143,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -203,7 +200,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -231,7 +227,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -245,8 +240,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
Outra opção útil para habilitar é retomar o treinamento de um checkpoint anterior. Isso garantirá que você possa continuar de onde parou sem recomeçar se o seu treinamento for interrompido. Existem dois métodos para retomar o treinamento a partir de um checkpoint.
|
||||
|
||||
O primeiro método usa o argumento `output_dir previous_output_dir` para retomar o treinamento do último checkpoint armazenado em `output_dir`. Neste caso, você deve remover `overwrite_output_dir`:
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -258,24 +251,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
O segundo método usa o argumento `resume_from_checkpoint path_to_specific_checkpoint` para retomar o treinamento de uma pasta de checkpoint específica.
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -307,6 +282,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -179,7 +179,7 @@ deepspeed --num_gpus=2 your_program.py <normal cl args> --deepspeed ds_config.js
|
||||
deepspeed examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero3.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -202,7 +202,7 @@ deepspeed examples/pytorch/translation/run_translation.py \
|
||||
deepspeed --num_gpus=1 examples/pytorch/translation/run_translation.py \
|
||||
--deepspeed tests/deepspeed/ds_config_zero2.json \
|
||||
--model_name_or_path google-t5/t5-small --per_device_train_batch_size 1 \
|
||||
--output_dir output_dir --overwrite_output_dir --fp16 \
|
||||
--output_dir output_dir --fp16 \
|
||||
--do_train --max_train_samples 500 --num_train_epochs 1 \
|
||||
--dataset_name wmt16 --dataset_config "ro-en" \
|
||||
--source_lang en --target_lang ro
|
||||
@ -1659,7 +1659,7 @@ deepspeed examples/pytorch/translation/run_translation.py \
|
||||
--model_name_or_path google-t5/t5-small --output_dir output_dir \
|
||||
--do_eval --max_eval_samples 50 --warmup_steps 50 \
|
||||
--max_source_length 128 --val_max_target_length 128 \
|
||||
--overwrite_output_dir --per_device_eval_batch_size 4 \
|
||||
--per_device_eval_batch_size 4 \
|
||||
--predict_with_generate --dataset_config "ro-en" --fp16 \
|
||||
--source_lang en --target_lang ro --dataset_name wmt16 \
|
||||
--source_prefix "translate English to Romanian: "
|
||||
|
||||
@ -471,7 +471,6 @@ python examples/pytorch/text-classification/run_glue.py \
|
||||
--learning_rate 2e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
**需要注意的一些注意事项**
|
||||
@ -606,7 +605,6 @@ accelerate launch \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
4. 你也可以直接使用`accelerate launch`的cmd参数。上面的示例将映射到：
|
||||
@ -631,7 +629,6 @@ accelerate launch --num_processes=2 \
|
||||
--learning_rate 5e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
有关更多信息，请参阅 🤗 Accelerate CLI 指南：[启动您的 🤗 Accelerate 脚本](https://huggingface.co/docs/accelerate/basic_tutorials/launch)。
|
||||
|
||||
@ -50,7 +50,6 @@ python examples/pytorch/text-classification/run_glue.py \
|
||||
--learning_rate 2e-5 \
|
||||
--num_train_epochs 3 \
|
||||
--output_dir /tmp/$TASK_NAME/ \
|
||||
--overwrite_output_dir
|
||||
```
|
||||
|
||||
用于[分布式设置](https://pytorch.org/docs/stable/distributed.html#backends)的后端(如`gloo`和`nccl`)不支持`mps`设备，这也意味着使用 MPS 后端时只能在单个 GPU 上进行训练。
|
||||
|
||||
@ -99,7 +99,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -124,7 +123,6 @@ torchrun \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -145,7 +143,6 @@ python xla_spawn.py --num_cores 8 \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -202,7 +199,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--summary_column summary_column_name \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
@ -231,7 +227,6 @@ python examples/pytorch/summarization/run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -245,8 +240,6 @@ examples/pytorch/summarization/run_summarization.py -h
|
||||
|
||||
另一个有用的选项是从之前的checkpoint恢复训练。这将确保在训练中断时，您可以从之前停止的地方继续进行，而无需重新开始。有两种方法可以从checkpoint恢复训练。
|
||||
|
||||
第一种方法使用`output_dir previous_output_dir`参数从存储在`output_dir`中的最新的checkpoint恢复训练。在这种情况下，您应该删除`overwrite_output_dir`：
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
@ -258,25 +251,6 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--output_dir previous_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
第二种方法使用`resume_from_checkpoint path_to_specific_checkpoint`参数从特定的checkpoint文件夹恢复训练。
|
||||
|
||||
|
||||
```bash
|
||||
python examples/pytorch/summarization/run_summarization.py
|
||||
--model_name_or_path google-t5/t5-small \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--dataset_name cnn_dailymail \
|
||||
--dataset_config "3.0.0" \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--resume_from_checkpoint path_to_specific_checkpoint \
|
||||
--predict_with_generate
|
||||
```
|
||||
@ -309,6 +283,5 @@ python examples/pytorch/summarization/run_summarization.py
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -95,17 +95,6 @@ def main():
|
||||
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
if (
|
||||
os.path.exists(training_args.output_dir)
|
||||
and os.listdir(training_args.output_dir)
|
||||
and training_args.do_train
|
||||
and not training_args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. Use"
|
||||
" --overwrite_output_dir to overcome."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
|
||||
@ -641,9 +641,6 @@ def main():
|
||||
help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number",
|
||||
)
|
||||
parser.add_argument("--no_cuda", action="store_true", help="Whether not to use CUDA when available")
|
||||
parser.add_argument(
|
||||
"--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets"
|
||||
)
|
||||
@ -677,18 +674,6 @@ def main():
|
||||
"stride or increase the maximum length to ensure the features are correctly built."
|
||||
)
|
||||
|
||||
if (
|
||||
os.path.exists(args.output_dir)
|
||||
and os.listdir(args.output_dir)
|
||||
and args.do_train
|
||||
and not args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(
|
||||
args.output_dir
|
||||
)
|
||||
)
|
||||
|
||||
# Setup distant debugging if needed
|
||||
if args.server_ip and args.server_port:
|
||||
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
|
||||
|
||||
@ -76,17 +76,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
if (
|
||||
os.path.exists(training_args.output_dir)
|
||||
and os.listdir(training_args.output_dir)
|
||||
and training_args.do_train
|
||||
and not training_args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. Use"
|
||||
" --overwrite_output_dir to overcome."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
|
||||
@ -207,16 +207,6 @@ def main():
|
||||
"Cannot do evaluation without an evaluation data file. Either supply a file to --eval_data_file "
|
||||
"or remove the --do_eval argument."
|
||||
)
|
||||
if (
|
||||
os.path.exists(training_args.output_dir)
|
||||
and os.listdir(training_args.output_dir)
|
||||
and training_args.do_train
|
||||
and not training_args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. Use"
|
||||
" --overwrite_output_dir to overcome."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
|
||||
@ -557,9 +557,6 @@ def main():
|
||||
help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number",
|
||||
)
|
||||
parser.add_argument("--no_cuda", action="store_true", help="Whether not to use CUDA when available")
|
||||
parser.add_argument(
|
||||
"--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets"
|
||||
)
|
||||
@ -584,18 +581,6 @@ def main():
|
||||
parser.add_argument("--server_port", type=str, default="", help="Can be used for distant debugging.")
|
||||
args = parser.parse_args()
|
||||
|
||||
if (
|
||||
os.path.exists(args.output_dir)
|
||||
and os.listdir(args.output_dir)
|
||||
and args.do_train
|
||||
and not args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(
|
||||
args.output_dir
|
||||
)
|
||||
)
|
||||
|
||||
# Setup distant debugging if needed
|
||||
if args.server_ip and args.server_port:
|
||||
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
|
||||
|
||||
@ -39,7 +39,6 @@ from utils import (
|
||||
Seq2SeqDataset,
|
||||
assert_all_frozen,
|
||||
build_compute_metrics_fn,
|
||||
check_output_dir,
|
||||
freeze_embeds,
|
||||
freeze_params,
|
||||
lmap,
|
||||
@ -168,8 +167,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
check_output_dir(training_args)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
|
||||
@ -20,7 +20,7 @@ export MAX_LEN=128
|
||||
python finetune_trainer.py \
|
||||
--tokenizer_name $m --model_name_or_path $m \
|
||||
--data_dir $ENRO_DIR \
|
||||
--output_dir marian_en_ro_6_3 --overwrite_output_dir \
|
||||
--output_dir marian_en_ro_6_3 \
|
||||
--learning_rate=3e-4 \
|
||||
--warmup_steps 500 --sortish_sampler \
|
||||
--fp16 \
|
||||
|
||||
@ -22,7 +22,7 @@ python xla_spawn.py --num_cores $TPU_NUM_CORES \
|
||||
finetune_trainer.py \
|
||||
--tokenizer_name $m --model_name_or_path $m \
|
||||
--data_dir $ENRO_DIR \
|
||||
--output_dir marian_en_ro_6_3 --overwrite_output_dir \
|
||||
--output_dir marian_en_ro_6_3 \
|
||||
--learning_rate=3e-4 \
|
||||
--warmup_steps 500 \
|
||||
--per_device_train_batch_size=$BS --per_device_eval_batch_size=$BS \
|
||||
|
||||
@ -21,7 +21,7 @@ export MAX_TGT_LEN=142
|
||||
python finetune_trainer.py \
|
||||
--model_name_or_path $m --tokenizer_name $tok \
|
||||
--data_dir cnn_dm \
|
||||
--output_dir distilbart-cnn-12-6 --overwrite_output_dir \
|
||||
--output_dir distilbart-cnn-12-6 \
|
||||
--learning_rate=3e-5 \
|
||||
--warmup_steps 500 --sortish_sampler \
|
||||
--fp16 \
|
||||
|
||||
@ -15,7 +15,7 @@
|
||||
python finetune_trainer.py \
|
||||
--model_name_or_path=facebook/mbart-large-cc25 \
|
||||
--data_dir $ENRO_DIR \
|
||||
--output_dir mbart_cc25_enro --overwrite_output_dir \
|
||||
--output_dir mbart_cc25_enro \
|
||||
--learning_rate=3e-5 \
|
||||
--warmup_steps 500 \
|
||||
--fp16 \
|
||||
|
||||
@ -639,27 +639,3 @@ def chunks(lst, n):
|
||||
"""Yield successive n-sized chunks from lst."""
|
||||
for i in range(0, len(lst), n):
|
||||
yield lst[i : i + n]
|
||||
|
||||
|
||||
def check_output_dir(args, expected_items=0):
|
||||
"""
|
||||
Checks whether to bail out if output_dir already exists and has more than expected_items in it
|
||||
|
||||
`args`: needs to have the following attributes of `args`:
|
||||
- output_dir
|
||||
- do_train
|
||||
- overwrite_output_dir
|
||||
|
||||
`expected_items`: normally 0 (default) - i.e. empty dir, but in some cases a few files are expected (e.g. recovery from OOM)
|
||||
"""
|
||||
if (
|
||||
os.path.exists(args.output_dir)
|
||||
and len(os.listdir(args.output_dir)) > expected_items
|
||||
and args.do_train
|
||||
and not args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
f"Output directory ({args.output_dir}) already exists and "
|
||||
f"has {len(os.listdir(args.output_dir))} items in it (expected {expected_items} items). "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
|
||||
@ -111,17 +111,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
if (
|
||||
os.path.exists(training_args.output_dir)
|
||||
and os.listdir(training_args.output_dir)
|
||||
and training_args.do_train
|
||||
and not training_args.overwrite_output_dir
|
||||
):
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. Use"
|
||||
" --overwrite_output_dir to overcome."
|
||||
)
|
||||
|
||||
module = import_module("tasks")
|
||||
try:
|
||||
token_classification_task_clazz = getattr(module, model_args.task_type)
|
||||
|
||||
@ -72,8 +72,7 @@ token-classification/run_ner.py -h
|
||||
|
||||
You can resume training from a previous checkpoint like this:
|
||||
|
||||
1. Pass `--output_dir previous_output_dir` without `--overwrite_output_dir` to resume training from the latest checkpoint in `output_dir` (what you would use if the training was interrupted, for instance).
|
||||
2. Pass `--resume_from_checkpoint path_to_a_specific_checkpoint` to resume training from that checkpoint folder.
|
||||
1. Pass `--resume_from_checkpoint path_to_a_specific_checkpoint` to resume training from that checkpoint folder.
|
||||
|
||||
Should you want to turn an example into a notebook where you'd no longer have access to the command
|
||||
line, 🤗 Trainer supports resuming from a checkpoint via `trainer.train(resume_from_checkpoint)`.
|
||||
|
||||
@ -34,7 +34,6 @@ python run_audio_classification.py \
|
||||
--dataset_name superb \
|
||||
--dataset_config_name ks \
|
||||
--output_dir wav2vec2-base-ft-keyword-spotting \
|
||||
--overwrite_output_dir \
|
||||
--remove_unused_columns False \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
@ -76,7 +75,6 @@ python run_audio_classification.py \
|
||||
--audio_column_name audio \
|
||||
--label_column_name language \
|
||||
--output_dir wav2vec2-base-lang-id \
|
||||
--overwrite_output_dir \
|
||||
--remove_unused_columns False \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
|
||||
@ -47,7 +47,6 @@ from transformers import (
|
||||
TrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -245,21 +244,6 @@ def main():
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to train from scratch."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Initialize our dataset and prepare it for the audio classification task.
|
||||
raw_datasets = DatasetDict()
|
||||
raw_datasets["train"] = load_dataset(
|
||||
@ -408,8 +392,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
|
||||
@ -97,6 +97,5 @@ python run_clip.py \
|
||||
--per_device_train_batch_size="64" \
|
||||
--per_device_eval_batch_size="64" \
|
||||
--learning_rate="5e-5" --warmup_steps="0" --weight_decay 0.1 \
|
||||
--overwrite_output_dir \
|
||||
--push_to_hub
|
||||
```
|
||||
|
||||
@ -55,7 +55,6 @@ from transformers import (
|
||||
TrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -271,21 +270,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# 3. Detecting last checkpoint and eventually continue from last checkpoint
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# 4. Load dataset
|
||||
# Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
|
||||
# or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
|
||||
@ -497,8 +481,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
tokenizer.save_pretrained(training_args.output_dir)
|
||||
|
||||
@ -58,7 +58,6 @@ from transformers import (
|
||||
TrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -225,21 +224,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -418,8 +402,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
|
||||
@ -44,7 +44,6 @@ Alternatively, one can decide to further pre-train an already pre-trained (or fi
|
||||
!python run_mim.py \
|
||||
--model_type vit \
|
||||
--output_dir ./outputs/ \
|
||||
--overwrite_output_dir \
|
||||
--remove_unused_columns False \
|
||||
--label_names bool_masked_pos \
|
||||
--do_train \
|
||||
@ -95,7 +94,6 @@ Next, we can run the script by providing the path to this custom configuration (
|
||||
--config_name_or_path path_to_config \
|
||||
--model_type swin \
|
||||
--output_dir ./outputs/ \
|
||||
--overwrite_output_dir \
|
||||
--remove_unused_columns False \
|
||||
--label_names bool_masked_pos \
|
||||
--do_train \
|
||||
|
||||
@ -41,7 +41,6 @@ from transformers import (
|
||||
ViTMAEConfig,
|
||||
ViTMAEForPreTraining,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -217,21 +216,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Initialize our dataset.
|
||||
ds = load_dataset(
|
||||
data_args.dataset_name,
|
||||
@ -377,8 +361,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
|
||||
@ -44,7 +44,6 @@ from transformers import (
|
||||
Trainer,
|
||||
TrainingArguments,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -281,21 +280,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Initialize our dataset.
|
||||
ds = load_dataset(
|
||||
data_args.dataset_name,
|
||||
@ -456,8 +440,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
|
||||
@ -49,7 +49,6 @@ from transformers import (
|
||||
)
|
||||
from transformers.image_processing_utils import BatchFeature
|
||||
from transformers.trainer import EvalPrediction
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -334,18 +333,6 @@ def find_last_checkpoint(training_args: TrainingArguments) -> Optional[str]:
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif os.path.isdir(training_args.output_dir) and not training_args.overwrite_output_dir:
|
||||
checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
return checkpoint
|
||||
|
||||
|
||||
@ -63,7 +63,6 @@ from transformers import (
|
||||
set_seed,
|
||||
)
|
||||
from transformers.testing_utils import CaptureLogger
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -317,21 +316,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -665,8 +649,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -66,7 +66,6 @@ from transformers import (
|
||||
)
|
||||
from transformers.integrations import is_deepspeed_zero3_enabled
|
||||
from transformers.testing_utils import CaptureLogger
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -344,21 +343,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -806,8 +790,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -62,7 +62,6 @@ from transformers import (
|
||||
is_torch_xla_available,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -290,21 +289,6 @@ def main():
|
||||
# Set the verbosity to info of the Transformers logger (on main process only):
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -631,8 +615,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
metrics = train_result.metrics
|
||||
|
||||
@ -55,7 +55,6 @@ from transformers import (
|
||||
XLNetLMHeadModel,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -269,21 +268,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -528,8 +512,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
metrics = train_result.metrics
|
||||
|
||||
@ -52,7 +52,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
|
||||
|
||||
@ -213,21 +212,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -401,8 +385,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
metrics = train_result.metrics
|
||||
|
||||
@ -51,7 +51,6 @@ from transformers import (
|
||||
from transformers.image_processing_utils import BatchFeature
|
||||
from transformers.image_transforms import center_to_corners_format
|
||||
from transformers.trainer import EvalPrediction
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -373,23 +372,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif os.path.isdir(training_args.output_dir) and not training_args.overwrite_output_dir:
|
||||
checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# ------------------------------------------------------------------------------------------------
|
||||
# Load dataset, prepare splits
|
||||
# ------------------------------------------------------------------------------------------------
|
||||
@ -510,7 +492,7 @@ def main():
|
||||
|
||||
# Training
|
||||
if training_args.do_train:
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
trainer.save_metrics("train", train_result.metrics)
|
||||
|
||||
@ -55,7 +55,6 @@ class TorchXLAExamplesTests(TestCasePlus):
|
||||
./examples/pytorch/text-classification/run_glue.py
|
||||
--model_name_or_path distilbert/distilbert-base-uncased
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--train_file ./tests/fixtures/tests_samples/MRPC/train.csv
|
||||
--validation_file ./tests/fixtures/tests_samples/MRPC/dev.csv
|
||||
--do_train
|
||||
|
||||
@ -43,7 +43,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -262,21 +261,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -646,8 +630,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -41,7 +41,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -260,21 +259,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -672,8 +656,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -39,7 +39,7 @@ from transformers import (
|
||||
Seq2SeqTrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import EvalLoopOutput, EvalPrediction, get_last_checkpoint
|
||||
from transformers.trainer_utils import EvalLoopOutput, EvalPrediction
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -307,21 +307,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -669,8 +654,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -52,7 +52,6 @@ from transformers import (
|
||||
TrainingArguments,
|
||||
default_data_collator,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -221,21 +220,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Load dataset
|
||||
# In distributed training, the load_dataset function guarantees that only one local process can concurrently
|
||||
# download the dataset.
|
||||
@ -419,8 +403,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model()
|
||||
trainer.log_metrics("train", train_result.metrics)
|
||||
|
||||
@ -70,7 +70,6 @@ python run_speech_recognition_ctc.py \
|
||||
--model_name_or_path="facebook/wav2vec2-large-xlsr-53" \
|
||||
--dataset_config_name="tr" \
|
||||
--output_dir="./wav2vec2-common_voice-tr-demo" \
|
||||
--overwrite_output_dir \
|
||||
--num_train_epochs="15" \
|
||||
--per_device_train_batch_size="16" \
|
||||
--gradient_accumulation_steps="2" \
|
||||
@ -106,7 +105,6 @@ torchrun \
|
||||
--model_name_or_path="facebook/wav2vec2-large-xlsr-53" \
|
||||
--dataset_config_name="tr" \
|
||||
--output_dir="./wav2vec2-common_voice-tr-demo-dist" \
|
||||
--overwrite_output_dir \
|
||||
--num_train_epochs="15" \
|
||||
--per_device_train_batch_size="4" \
|
||||
--learning_rate="3e-4" \
|
||||
@ -156,7 +154,6 @@ However, the `--shuffle_buffer_size` argument controls how many examples we can
|
||||
--train_split_name="train+validation" \
|
||||
--eval_split_name="test" \
|
||||
--output_dir="wav2vec2-xls-r-common_voice-tr-ft" \
|
||||
--overwrite_output_dir \
|
||||
--max_steps="5000" \
|
||||
--per_device_train_batch_size="8" \
|
||||
--gradient_accumulation_steps="2" \
|
||||
@ -390,7 +387,6 @@ python run_speech_recognition_seq2seq.py \
|
||||
--freeze_feature_encoder="False" \
|
||||
--gradient_checkpointing \
|
||||
--fp16 \
|
||||
--overwrite_output_dir \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--predict_with_generate \
|
||||
@ -431,7 +427,6 @@ torchrun \
|
||||
--freeze_feature_encoder="False" \
|
||||
--gradient_checkpointing \
|
||||
--fp16 \
|
||||
--overwrite_output_dir \
|
||||
--do_train \
|
||||
--do_eval \
|
||||
--predict_with_generate \
|
||||
@ -539,7 +534,6 @@ python run_speech_recognition_seq2seq.py \
|
||||
--output_dir="./" \
|
||||
--preprocessing_num_workers="16" \
|
||||
--length_column_name="input_length" \
|
||||
--overwrite_output_dir \
|
||||
--num_train_epochs="5" \
|
||||
--per_device_train_batch_size="8" \
|
||||
--per_device_eval_batch_size="8" \
|
||||
@ -581,7 +575,6 @@ torchrun \
|
||||
--output_dir="./" \
|
||||
--preprocessing_num_workers="16" \
|
||||
--length_column_name="input_length" \
|
||||
--overwrite_output_dir \
|
||||
--num_train_epochs="5" \
|
||||
--per_device_train_batch_size="8" \
|
||||
--per_device_eval_batch_size="8" \
|
||||
|
||||
@ -55,7 +55,7 @@ from transformers import (
|
||||
Wav2Vec2Processor,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint, is_main_process
|
||||
from transformers.trainer_utils import is_main_process
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -429,21 +429,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
@ -558,7 +543,7 @@ def main():
|
||||
vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
|
||||
|
||||
with training_args.main_process_first():
|
||||
if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
|
||||
if os.path.isfile(vocab_file):
|
||||
try:
|
||||
os.remove(vocab_file)
|
||||
except OSError:
|
||||
@ -781,9 +766,7 @@ def main():
|
||||
# Training
|
||||
if training_args.do_train:
|
||||
# use last checkpoint if exist
|
||||
if last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
elif os.path.isdir(model_args.model_name_or_path):
|
||||
if os.path.isdir(model_args.model_name_or_path):
|
||||
checkpoint = model_args.model_name_or_path
|
||||
else:
|
||||
checkpoint = None
|
||||
|
||||
@ -58,7 +58,7 @@ from transformers import (
|
||||
set_seed,
|
||||
)
|
||||
from transformers.models.wav2vec2.modeling_wav2vec2 import WAV2VEC2_ADAPTER_SAFE_FILE
|
||||
from transformers.trainer_utils import get_last_checkpoint, is_main_process
|
||||
from transformers.trainer_utils import is_main_process
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -406,21 +406,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
@ -557,7 +542,7 @@ def main():
|
||||
vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
|
||||
|
||||
with training_args.main_process_first():
|
||||
if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
|
||||
if os.path.isfile(vocab_file):
|
||||
try:
|
||||
os.remove(vocab_file)
|
||||
except OSError:
|
||||
@ -773,9 +758,7 @@ def main():
|
||||
# Training
|
||||
if training_args.do_train:
|
||||
# use last checkpoint if exist
|
||||
if last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
elif os.path.isdir(model_args.model_name_or_path):
|
||||
if os.path.isdir(model_args.model_name_or_path):
|
||||
checkpoint = model_args.model_name_or_path
|
||||
else:
|
||||
checkpoint = None
|
||||
|
||||
@ -54,7 +54,7 @@ from transformers import (
|
||||
Seq2SeqTrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint, is_main_process
|
||||
from transformers.trainer_utils import is_main_process
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -340,21 +340,6 @@ def main():
|
||||
transformers.utils.logging.set_verbosity_info()
|
||||
logger.info("Training/evaluation parameters %s", training_args)
|
||||
|
||||
# 3. Detecting last checkpoint and eventually continue from last checkpoint
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -603,8 +588,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the feature extractor too for easy upload
|
||||
|
||||
|
||||
@ -50,7 +50,6 @@ python run_summarization.py \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -72,7 +71,6 @@ python run_summarization.py \
|
||||
--validation_file path_to_csv_or_jsonlines_file \
|
||||
--source_prefix "summarize: " \
|
||||
--output_dir /tmp/tst-summarization \
|
||||
--overwrite_output_dir \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--predict_with_generate
|
||||
|
||||
@ -61,7 +61,6 @@ from transformers import (
|
||||
Seq2SeqTrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version, is_offline_mode
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -374,21 +373,6 @@ def main():
|
||||
"`--source_prefix 'summarize: ' `"
|
||||
)
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -698,8 +682,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -104,7 +104,6 @@ class ExamplesTests(TestCasePlus):
|
||||
run_glue.py
|
||||
--model_name_or_path distilbert/distilbert-base-uncased
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--train_file ./tests/fixtures/tests_samples/MRPC/train.csv
|
||||
--validation_file ./tests/fixtures/tests_samples/MRPC/dev.csv
|
||||
--do_train
|
||||
@ -140,7 +139,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_eval_batch_size 5
|
||||
--num_train_epochs 2
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
""".split()
|
||||
|
||||
if backend_device_count(torch_device) > 1:
|
||||
@ -188,7 +186,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file ./tests/fixtures/sample_text.txt
|
||||
--validation_file ./tests/fixtures/sample_text.txt
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--do_train
|
||||
--do_eval
|
||||
--prediction_loss_only
|
||||
@ -214,7 +211,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/conll/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/conll/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--do_train
|
||||
--do_eval
|
||||
--warmup_steps=2
|
||||
@ -243,7 +239,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/SQUAD/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/SQUAD/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--max_steps=10
|
||||
--warmup_steps=2
|
||||
--do_train
|
||||
@ -271,7 +266,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/SQUAD/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/SQUAD/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--max_steps=10
|
||||
--warmup_steps=2
|
||||
--do_train
|
||||
@ -296,7 +290,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/swag/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/swag/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--max_steps=20
|
||||
--warmup_steps=2
|
||||
--do_train
|
||||
@ -334,7 +327,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/xsum/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/xsum/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--max_steps=50
|
||||
--warmup_steps=8
|
||||
--do_train
|
||||
@ -364,7 +356,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--train_file tests/fixtures/tests_samples/wmt16/sample.json
|
||||
--validation_file tests/fixtures/tests_samples/wmt16/sample.json
|
||||
--output_dir {tmp_dir}
|
||||
--overwrite_output_dir
|
||||
--max_steps=50
|
||||
--warmup_steps=8
|
||||
--do_train
|
||||
@ -396,7 +387,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 1
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--dataloader_num_workers 16
|
||||
--metric_for_best_model accuracy
|
||||
--max_steps 10
|
||||
@ -429,7 +419,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 1
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--preprocessing_num_workers 16
|
||||
--max_steps 10
|
||||
--seed 42
|
||||
@ -459,7 +448,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 1
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--preprocessing_num_workers 16
|
||||
--max_steps 10
|
||||
--target_language tur
|
||||
@ -491,7 +479,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 4
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--preprocessing_num_workers 16
|
||||
--max_steps 10
|
||||
--seed 42
|
||||
@ -523,7 +510,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 1
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--num_train_epochs 10
|
||||
--max_steps 50
|
||||
--seed 42
|
||||
@ -572,7 +558,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--per_device_train_batch_size 2
|
||||
--per_device_eval_batch_size 1
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--dataloader_num_workers 16
|
||||
--metric_for_best_model accuracy
|
||||
--max_steps 10
|
||||
@ -597,7 +582,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--do_train
|
||||
--do_eval
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--max_steps 10
|
||||
--learning_rate=2e-4
|
||||
--per_device_train_batch_size=2
|
||||
@ -624,7 +608,6 @@ class ExamplesTests(TestCasePlus):
|
||||
--do_train
|
||||
--do_eval
|
||||
--remove_unused_columns False
|
||||
--overwrite_output_dir True
|
||||
--eval_do_concat_batches False
|
||||
--max_steps 10
|
||||
--learning_rate=1e-6
|
||||
|
||||
@ -55,7 +55,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -321,21 +320,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -693,8 +677,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
metrics = train_result.metrics
|
||||
max_train_samples = (
|
||||
|
||||
@ -57,7 +57,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -266,21 +265,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -566,8 +550,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
metrics = train_result.metrics
|
||||
max_train_samples = (
|
||||
|
||||
@ -56,7 +56,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -224,21 +223,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -412,8 +396,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
metrics = train_result.metrics
|
||||
max_train_samples = (
|
||||
|
||||
@ -54,7 +54,6 @@ from transformers import (
|
||||
TrainingArguments,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -263,21 +262,6 @@ def main():
|
||||
)
|
||||
logger.info(f"Training/evaluation parameters {training_args}")
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -587,8 +571,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
metrics = train_result.metrics
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
@ -53,7 +53,6 @@ python run_translation.py \
|
||||
--output_dir /tmp/tst-translation \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -74,7 +73,6 @@ python run_translation.py \
|
||||
--output_dir /tmp/tst-translation \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -96,7 +94,6 @@ python run_translation.py \
|
||||
--output_dir /tmp/tst-translation \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -118,7 +115,6 @@ python run_translation.py \
|
||||
--output_dir /tmp/tst-translation \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
@ -144,7 +140,6 @@ python run_translation.py \
|
||||
--output_dir /tmp/tst-translation \
|
||||
--per_device_train_batch_size=4 \
|
||||
--per_device_eval_batch_size=4 \
|
||||
--overwrite_output_dir \
|
||||
--predict_with_generate
|
||||
```
|
||||
|
||||
|
||||
@ -60,7 +60,6 @@ from transformers import (
|
||||
default_data_collator,
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.utils import check_min_version
|
||||
from transformers.utils.versions import require_version
|
||||
|
||||
@ -322,21 +321,6 @@ def main():
|
||||
"`--source_prefix 'translate English to German: ' `"
|
||||
)
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Set seed before initializing model.
|
||||
set_seed(training_args.seed)
|
||||
|
||||
@ -617,8 +601,6 @@ def main():
|
||||
checkpoint = None
|
||||
if training_args.resume_from_checkpoint is not None:
|
||||
checkpoint = training_args.resume_from_checkpoint
|
||||
elif last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
train_result = trainer.train(resume_from_checkpoint=checkpoint)
|
||||
trainer.save_model() # Saves the tokenizer too for easy upload
|
||||
|
||||
|
||||
@ -1644,7 +1644,7 @@ class NeptuneCallback(TrainerCallback):
|
||||
|
||||
def on_init_end(self, args, state, control, **kwargs):
|
||||
self._volatile_checkpoints_dir = None
|
||||
if self._log_checkpoints and (args.overwrite_output_dir or args.save_total_limit is not None):
|
||||
if self._log_checkpoints and args.save_total_limit is not None:
|
||||
self._volatile_checkpoints_dir = tempfile.TemporaryDirectory().name
|
||||
|
||||
if self._log_checkpoints == "best" and not args.load_best_model_at_end:
|
||||
|
||||
@ -21,11 +21,11 @@
|
||||
|
||||
from typing import Optional
|
||||
|
||||
from ...configuration_utils import PretrainedConfig, layer_type_validation
|
||||
from ...configuration_utils import PreTrainedConfig, layer_type_validation
|
||||
from ...modeling_rope_utils import rope_config_validation
|
||||
|
||||
|
||||
class CwmConfig(PretrainedConfig):
|
||||
class CwmConfig(PreTrainedConfig):
|
||||
"""
|
||||
Configuration for Code World Model (CWM).
|
||||
This is an inherited Llama3-compatible configuration with layer-interleaved
|
||||
@ -136,13 +136,6 @@ class CwmConfig(PretrainedConfig):
|
||||
layer_types: Optional[list[str]] = None, # ["full_attention"|"sliding_attention"] per layer
|
||||
**kwargs,
|
||||
):
|
||||
super().__init__(
|
||||
pad_token_id=pad_token_id,
|
||||
bos_token_id=bos_token_id,
|
||||
eos_token_id=eos_token_id,
|
||||
tie_word_embeddings=tie_word_embeddings,
|
||||
**kwargs,
|
||||
)
|
||||
if rope_scaling is None:
|
||||
rope_scaling = {
|
||||
"factor": 16.0,
|
||||
@ -189,6 +182,14 @@ class CwmConfig(PretrainedConfig):
|
||||
self.rope_scaling["rope_type"] = self.rope_scaling["type"]
|
||||
rope_config_validation(self)
|
||||
|
||||
super().__init__(
|
||||
pad_token_id=pad_token_id,
|
||||
bos_token_id=bos_token_id,
|
||||
eos_token_id=eos_token_id,
|
||||
tie_word_embeddings=tie_word_embeddings,
|
||||
**kwargs,
|
||||
)
|
||||
|
||||
self.sliding_window = int(sliding_window) if sliding_window else None
|
||||
self.layer_types = list(layer_types)
|
||||
|
||||
|
||||
@ -19,7 +19,8 @@
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
from typing import Callable, Optional, Union
|
||||
from collections.abc import Callable
|
||||
from typing import Optional, Union
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
@ -36,7 +37,6 @@ from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
|
||||
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
|
||||
from ...processing_utils import Unpack
|
||||
from ...utils import TransformersKwargs, auto_docstring, can_return_tuple
|
||||
from ...utils.deprecation import deprecate_kwarg
|
||||
from ...utils.generic import check_model_inputs
|
||||
from .configuration_cwm import CwmConfig
|
||||
|
||||
@ -131,7 +131,6 @@ class CwmAttention(nn.Module):
|
||||
self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
|
||||
self.sliding_window = config.sliding_window if config.layer_types[layer_idx] == "sliding_attention" else None
|
||||
|
||||
@deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
|
||||
def forward(
|
||||
self,
|
||||
hidden_states: torch.Tensor,
|
||||
@ -225,7 +224,6 @@ class CwmDecoderLayer(GradientCheckpointingLayer):
|
||||
self.post_attention_layernorm = CwmRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
||||
self.attention_type = config.layer_types[layer_idx]
|
||||
|
||||
@deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
|
||||
def forward(
|
||||
self,
|
||||
hidden_states: torch.Tensor,
|
||||
@ -339,7 +337,7 @@ class CwmModel(CwmPreTrainedModel):
|
||||
# Initialize weights and apply final processing
|
||||
self.post_init()
|
||||
|
||||
@check_model_inputs
|
||||
@check_model_inputs()
|
||||
@auto_docstring
|
||||
def forward(
|
||||
self,
|
||||
|
||||
@ -208,9 +208,6 @@ class TrainingArguments:
|
||||
Parameters:
|
||||
output_dir (`str`, *optional*, defaults to `"trainer_output"`):
|
||||
The output directory where the model predictions and checkpoints will be written.
|
||||
overwrite_output_dir (`bool`, *optional*, defaults to `False`):
|
||||
If `True`, overwrite the content of the output directory. Use this to continue training if `output_dir`
|
||||
points to a checkpoint directory.
|
||||
do_train (`bool`, *optional*, defaults to `False`):
|
||||
Whether to run training or not. This argument is not directly used by [`Trainer`], it's intended to be used
|
||||
by your training/evaluation scripts instead. See the [example
|
||||
@ -787,15 +784,6 @@ class TrainingArguments:
|
||||
"help": "The output directory where the model predictions and checkpoints will be written. Defaults to 'trainer_output' if not provided."
|
||||
},
|
||||
)
|
||||
overwrite_output_dir: bool = field(
|
||||
default=False,
|
||||
metadata={
|
||||
"help": (
|
||||
"Overwrite the content of the output directory. "
|
||||
"Use this to continue training if output_dir points to a checkpoint directory."
|
||||
)
|
||||
},
|
||||
)
|
||||
|
||||
do_train: bool = field(default=False, metadata={"help": "Whether to run training."})
|
||||
do_eval: bool = field(default=False, metadata={"help": "Whether to run eval on the dev set."})
|
||||
|
||||
@ -219,21 +219,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
@ -452,16 +437,12 @@ def main():
|
||||
# Training
|
||||
if training_args.do_train:
|
||||
{%- if cookiecutter.can_train_from_scratch == "False" %}
|
||||
if last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
elif os.path.isdir(model_args.model_name_or_path):
|
||||
if os.path.isdir(model_args.model_name_or_path):
|
||||
checkpoint = model_args.model_name_or_path
|
||||
else:
|
||||
checkpoint = None
|
||||
{%- elif cookiecutter.can_train_from_scratch == "True" %}
|
||||
if last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
|
||||
if model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
|
||||
checkpoint = model_args.model_name_or_path
|
||||
else:
|
||||
checkpoint = None
|
||||
|
||||
@ -1303,7 +1303,6 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
|
||||
--train_file {data_dir}/train.json
|
||||
--validation_file {data_dir}/val.json
|
||||
--output_dir {output_dir}
|
||||
--overwrite_output_dir
|
||||
--max_source_length {max_len}
|
||||
--max_target_length {max_len}
|
||||
--val_max_target_length {max_len}
|
||||
@ -1373,7 +1372,6 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
|
||||
--train_file {data_dir}/sample_text.txt
|
||||
--validation_file {data_dir}/sample_text.txt
|
||||
--output_dir {output_dir}
|
||||
--overwrite_output_dir
|
||||
--do_train
|
||||
--do_eval
|
||||
--max_train_samples 16
|
||||
@ -1410,7 +1408,6 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
|
||||
--train_file {data_dir}/sample_text.txt
|
||||
--validation_file {data_dir}/sample_text.txt
|
||||
--output_dir {output_dir}
|
||||
--overwrite_output_dir
|
||||
--do_train
|
||||
--max_train_samples 4
|
||||
--per_device_train_batch_size 2
|
||||
|
||||
@ -161,7 +161,6 @@ def make_task_cmds():
|
||||
--num_train_epochs 1
|
||||
--fp16
|
||||
--report_to none
|
||||
--overwrite_output_dir
|
||||
""".split()
|
||||
|
||||
# try to cover as many models as possible once (it's enough to run on one task per model)
|
||||
|
||||
@ -267,7 +267,6 @@ class TestTrainerExt(TestCasePlus):
|
||||
--validation_file {data_dir}/val.json
|
||||
--test_file {data_dir}/test.json
|
||||
--output_dir {output_dir}
|
||||
--overwrite_output_dir
|
||||
--max_train_samples 8
|
||||
--max_source_length {max_len}
|
||||
--max_target_length {max_len}
|
||||
|
||||
@ -447,7 +447,6 @@ class TrainerIntegrationFSDP(TestCasePlus, TrainerIntegrationCommon):
|
||||
--model_name_or_path google-bert/bert-base-cased
|
||||
--task_name mrpc
|
||||
--output_dir {output_dir}
|
||||
--overwrite_output_dir
|
||||
--do_train
|
||||
--max_seq_length 128
|
||||
--per_device_train_batch_size 16
|
||||
|
||||
@ -21,7 +21,6 @@ class SageMakerTestEnvironment:
|
||||
"do_eval": True,
|
||||
"do_predict": True,
|
||||
"output_dir": "/opt/ml/model",
|
||||
"overwrite_output_dir": True,
|
||||
"max_steps": 500,
|
||||
"save_steps": 5500,
|
||||
}
|
||||
|
||||
@ -38,7 +38,6 @@ from transformers import (
|
||||
set_seed,
|
||||
)
|
||||
from transformers.trainer import Trainer
|
||||
from transformers.trainer_utils import get_last_checkpoint
|
||||
from transformers.training_args import TrainingArguments
|
||||
from transformers.utils import check_min_version
|
||||
|
||||
@ -198,21 +197,6 @@ def main():
|
||||
else:
|
||||
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
|
||||
|
||||
# Detecting last checkpoint.
|
||||
last_checkpoint = None
|
||||
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
|
||||
last_checkpoint = get_last_checkpoint(training_args.output_dir)
|
||||
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
|
||||
raise ValueError(
|
||||
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
|
||||
"Use --overwrite_output_dir to overcome."
|
||||
)
|
||||
elif last_checkpoint is not None:
|
||||
logger.info(
|
||||
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
|
||||
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
|
||||
)
|
||||
|
||||
# Setup logging
|
||||
logging.basicConfig(
|
||||
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
||||
@ -459,9 +443,7 @@ def main():
|
||||
# Training
|
||||
if training_args.do_train:
|
||||
checkpoint = None
|
||||
if last_checkpoint is not None:
|
||||
checkpoint = last_checkpoint
|
||||
elif os.path.isdir(model_args.model_name_or_path):
|
||||
if os.path.isdir(model_args.model_name_or_path):
|
||||
# Check the config from that potential checkpoint has the right number of labels before using it as a
|
||||
# checkpoint.
|
||||
if AutoConfig.from_pretrained(model_args.model_name_or_path).num_labels == num_labels:
|
||||
|
||||
@ -4455,7 +4455,6 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
|
||||
"1",
|
||||
"--output_dir",
|
||||
tmpdir,
|
||||
"--overwrite_output_dir",
|
||||
"--do_train",
|
||||
"--max_train_samples",
|
||||
"64",
|
||||
|
||||
Reference in New Issue
Block a user