Hugging Face Transformers

os.environ["WANDB_PROJECT"] = "<my-project-name>"  # W&B 프로젝트 이름을 지정하세요
os.environ["WANDB_LOG_MODEL"] = "checkpoint"  # 모든 모델 checkpoint를 로깅하세요

from transformers import TrainingArguments, Trainer

args = TrainingArguments(..., report_to="wandb")  # W&B 로깅을 활성화하세요
trainer = Trainer(..., args=args)

run = wandb.init()
trainer.train()  # 트레이닝 및 W&B에 로깅 시작

# Post-training 분석, 테스트, 기타 로깅된 코드

run.finish()

# 새 run을 생성하세요
with wandb.init(project="amazon_sentiment_analysis") as run:
    # 아티팩트의 이름과 버전을 전달하세요
    my_model_name = "model-bert-base-high-lr:latest"
    my_model_artifact = run.use_artifact(my_model_name)

    # 모델 가중치를 폴더에 다운로드하고 경로를 반환하세요
    model_dir = my_model_artifact.download()

    # 해당 폴더에서 Hugging Face 모델을 불러오세요
    #  동일한 모델 클래스를 사용하세요
    model = AutoModelForSequenceClassification.from_pretrained(
        model_dir, num_labels=num_labels
    )

    # 추가 트레이닝을 수행하거나 추론을 실행하세요

last_run_id = "xxxxxxxx"  # wandb 작업 공간에서 run_id를 가져오세요

# run_id로 wandb run을 재개하세요
with wandb.init(
    project=os.environ["WANDB_PROJECT"],
    id=last_run_id,
    resume="must",
) as run:
    # run에 아티팩트를 연결하세요
    my_checkpoint_name = f"checkpoint-{last_run_id}:latest"
    my_checkpoint_artifact = run.use_artifact(my_checkpoint_name)

    # 체크포인트를 폴더에 다운로드하고 경로를 반환하세요
    checkpoint_dir = my_checkpoint_artifact.download()

    # 모델과 트레이너를 다시 초기화하세요
    model = AutoModelForSequenceClassification.from_pretrained(
        "<model-name>", num_labels=num_labels
    )
    # 여기에 트레이닝 인수를 지정하세요.
    training_args = TrainingArguments()

    trainer = Trainer(model=model, args=training_args)

    # 체크포인트에서 트레이닝을 재개하려면 반드시 체크포인트 디렉터리를 사용하세요
    trainer.train(resume_from_checkpoint=checkpoint_dir)

# 평소처럼 Trainer 인스턴스를 생성하세요
trainer = Trainer()

# Trainer 객체를 전달하여 새 로깅 callback 인스턴스를 생성하세요
evals_callback = WandbEvalsCallback(trainer, tokenizer, ...)

# Trainer에 callback을 추가하세요
trainer.add_callback(evals_callback)

# 평소처럼 Trainer 트레이닝을 시작하세요
trainer.train()

from transformers.integrations import WandbCallback
import pandas as pd


def decode_predictions(tokenizer, predictions):
    labels = tokenizer.batch_decode(predictions.label_ids)
    logits = predictions.predictions.argmax(axis=-1)
    prediction_text = tokenizer.batch_decode(logits)
    return {"labels": labels, "predictions": prediction_text}


class WandbPredictionProgressCallback(WandbCallback):
    """Custom WandbCallback to log model predictions during training.

    This callback logs model predictions and labels to a wandb.Table at each
    logging step during training. It allows to visualize the
    model predictions as the training progresses.

    Attributes:
        trainer (Trainer): The Hugging Face Trainer instance.
        tokenizer (AutoTokenizer): The tokenizer associated with the model.
        sample_dataset (Dataset): A subset of the validation dataset
          for generating predictions.
        num_samples (int, optional): Number of samples to select from
          the validation dataset for generating predictions. Defaults to 100.
        freq (int, optional): Frequency of logging. Defaults to 2.
    """

    def __init__(self, trainer, tokenizer, val_dataset, num_samples=100, freq=2):
        """Initializes the WandbPredictionProgressCallback instance.

        Args:
            trainer (Trainer): The Hugging Face Trainer instance.
            tokenizer (AutoTokenizer): The tokenizer associated
              with the model.
            val_dataset (Dataset): The validation dataset.
            num_samples (int, optional): Number of samples to select from
              the validation dataset for generating predictions.
              Defaults to 100.
            freq (int, optional): Frequency of logging. Defaults to 2.
        """
        super().__init__()
        self.trainer = trainer
        self.tokenizer = tokenizer
        self.sample_dataset = val_dataset.select(range(num_samples))
        self.freq = freq

    def on_evaluate(self, args, state, control, **kwargs):
        super().on_evaluate(args, state, control, **kwargs)
        # `freq` 에포크마다 예측을 로깅하여
        # 로깅 빈도를 제어합니다
        if state.epoch % self.freq == 0:
            # 예측을 생성합니다
            predictions = self.trainer.predict(self.sample_dataset)
            # 예측과 레이블을 디코딩합니다
            predictions = decode_predictions(self.tokenizer, predictions)
            # wandb.Table에 예측을 추가합니다
            predictions_df = pd.DataFrame(predictions)
            predictions_df["epoch"] = state.epoch
            records_table = self._wandb.Table(dataframe=predictions_df)
            # wandb에 table을 로깅합니다
            self._wandb.log({"sample_predictions": records_table})


# 먼저 Trainer를 인스턴스화합니다
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=lm_datasets["train"],
    eval_dataset=lm_datasets["validation"],
)

# WandbPredictionProgressCallback을 인스턴스화합니다
progress_callback = WandbPredictionProgressCallback(
    trainer=trainer,
    tokenizer=tokenizer,
    val_dataset=lm_dataset["validation"],
    num_samples=10,
    freq=2,
)

# trainer에 callback을 추가합니다
trainer.add_callback(progress_callback)

wandb.init(
    project="amazon_sentiment_analysis",
    name="bert-base-high-lr",
    tags=["baseline", "high-lr"],
    group="bert",
)