transformers

b96f213f - Clarify per_device_train_batch_size scaling in TrainingArguments (#38… (#38857)