I found that the validation of the eval dataset should be done in advance. It only prompted that the validation set was empty after I had trained for half an hour and completed one epoch.The code is as follows.I set eval_strategy="epoch" in train_args, but I don't set eval dataset in Trainer init.
System Info
I found that the validation of the eval dataset should be done in advance. It only prompted that the validation set was empty after I had trained for half an hour and completed one epoch.The code is as follows.I set eval_strategy="epoch" in train_args, but I don't set eval dataset in Trainer init.
`from transformers import TrainingArguments
train_args = TrainingArguments(output_dir="./checkpoints",
per_device_train_batch_size=2, per_device_eval_batch_size=1,
logging_steps=1,
eval_strategy="epoch",
save_strategy="epoch",
save_total_limit=3,
learning_rate=2e-5,
weight_decay=0.01, num_train_epochs=30, dataloader_drop_last=True, metric_for_best_model="f1",
load_best_model_at_end=True )
from transformers import DataCollatorWithPadding, Trainer
trainer = Trainer(model=model, args=train_args, train_dataset=process_dataset["train"], eval_dataset=None, data_collator=collate_fn)
`
transformers version:4.45.0 python version:3.10 paltform:ubuntu
@muellerzr @Sunm
Who can help?
No response
Information
Tasks
examplesfolder (such as GLUE/SQuAD, ...)Reproduction
Expected behavior
validation of the eval dataset should be done in advance