训练

train_model(experiment_name, experiment_dir, config, data_module, n_steps_train, metric_tracker=None, tokenizer=get_tokenizer(), peft=None, _use_rich_model_summary=True)

使用 PyTorch Lightning 训练 BioNeMo ESM2 模型。

参数

名称	类型	描述	默认值
experiment_name	str	实验的名称。	必需
experiment_dir	Path	实验将保存的目录。	必需
config	ESM2GenericConfig	ESM2 模型的配置。	必需
data_module	LightningDataModule	用于训练和验证的数据模块。	必需
n_steps_train	int	训练步骤的数量。	必需
metric_tracker	Callback | None	用于跟踪指标的可选回调	None
tokenizer	BioNeMoESMTokenizer	要使用的分词器。默认为 get_tokenizer()。	get_tokenizer()
peft	PEFT | None	PEFT（参数高效微调）模块。默认为 None。	None
_use_rich_model_summary	bool	是否使用 RichModelSummary 回调，在 https://nvbugspro.nvidia.com/bug/4959776 解决之前，我们的测试套件中省略了此回调。默认为 True。	True

返回值

类型	描述
Path	一个元组，包含保存的检查点路径、一个 MetricTracker
Callback | None	对象和 PyTorch Lightning Trainer 对象。

源代码位于 bionemo/esm2/model/finetune/train.py

def train_model(
    experiment_name: str,
    experiment_dir: Path,
    config: ESM2GenericConfig,
    data_module: pl.LightningDataModule,
    n_steps_train: int,
    metric_tracker: Callback | None = None,
    tokenizer: BioNeMoESMTokenizer = get_tokenizer(),
    peft: PEFT | None = None,
    _use_rich_model_summary: bool = True,
) -> Tuple[Path, Callback | None, nl.Trainer]:
    """Trains a BioNeMo ESM2 model using PyTorch Lightning.

    Parameters:
        experiment_name: The name of the experiment.
        experiment_dir: The directory where the experiment will be saved.
        config: The configuration for the ESM2 model.
        data_module: The data module for training and validation.
        n_steps_train: The number of training steps.
        metric_tracker: Optional callback to track metrics
        tokenizer: The tokenizer to use. Defaults to `get_tokenizer()`.
        peft: The PEFT (Parameter-Efficient Fine-Tuning) module. Defaults to None.
        _use_rich_model_summary: Whether to use the RichModelSummary callback, omitted in our test suite until
            https://nvbugspro.nvidia.com/bug/4959776 is resolved. Defaults to True.

    Returns:
        A tuple containing the path to the saved checkpoint, a MetricTracker
        object, and the PyTorch Lightning Trainer object.
    """
    checkpoint_callback = nl_callbacks.ModelCheckpoint(
        save_last=True,
        save_on_train_epoch_end=True,
        monitor="reduced_train_loss",  # TODO find out how to get val_loss logged and use "val_loss",
        every_n_train_steps=n_steps_train // 2,
        always_save_context=True,  # Enables the .nemo file-like checkpointing where all IOMixins are under SerDe
    )

    # Setup the logger and train the model
    nemo_logger = NeMoLogger(
        log_dir=str(experiment_dir),
        name=experiment_name,
        tensorboard=TensorBoardLogger(save_dir=experiment_dir, name=experiment_name),
        ckpt=checkpoint_callback,
    )
    # Needed so that the trainer can find an output directory for the profiler
    # ckpt_path needs to be a string for SerDe
    optimizer = MegatronOptimizerModule(
        config=OptimizerConfig(
            lr=5e-4,
            optimizer="adam",
            use_distributed_optimizer=True,
            fp16=config.fp16,
            bf16=config.bf16,
        )
    )
    module = biobert_lightning_module(config=config, tokenizer=tokenizer, optimizer=optimizer, model_transform=peft)

    strategy = nl.MegatronStrategy(
        tensor_model_parallel_size=1,
        pipeline_model_parallel_size=1,
        ddp="megatron",
        find_unused_parameters=True,
        enable_nemo_ckpt_io=True,
    )

    if _use_rich_model_summary:
        # RichModelSummary is not used in the test suite until https://nvbugspro.nvidia.com/bug/4959776 is resolved due
        # to errors with serialization / deserialization.
        callbacks: list[Callback] = [RichModelSummary(max_depth=4)]
    else:
        callbacks = []

    if metric_tracker is not None:
        callbacks.append(metric_tracker)
    if peft is not None:
        callbacks.append(
            ModelTransform()
        )  # Callback needed for PEFT fine-tuning using NeMo2, i.e. biobert_lightning_module(model_transform=peft).

    trainer = nl.Trainer(
        accelerator="gpu",
        devices=1,
        strategy=strategy,
        limit_val_batches=2,
        val_check_interval=n_steps_train // 2,
        max_steps=n_steps_train,
        num_nodes=1,
        log_every_n_steps=n_steps_train // 2,
        callbacks=callbacks,
        plugins=nl.MegatronMixedPrecision(precision="bf16-mixed"),
    )
    nllm.train(
        model=module,
        data=data_module,
        trainer=trainer,
        log=nemo_logger,
        resume=resume.AutoResume(
            resume_if_exists=True,  # Looks for the -last checkpoint to continue training.
            resume_ignore_no_checkpoint=True,  # When false this will throw an error with no existing checkpoint.
        ),
    )
    ckpt_path = Path(checkpoint_callback.last_model_path.replace(".ckpt", ""))
    return ckpt_path, metric_tracker, trainer