huggingface/transformers 用Trainer 和 不用Trainer

用Trainer：

from transformers import EvalPrediction, Trainer, TrainingArguments

training_args = TrainingArguments(
    output_dir=checkpoint_local,
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    weight_decay=0.01,
    save_strategy=
1
2
3
4
5
6
7