Infer esm2

get_parser()

返回此工具的 cli 解析器。

源代码位于 bionemo/esm2/scripts/infer_esm2.py

def get_parser():
    """Return the cli parser for this tool."""
    parser = argparse.ArgumentParser(description="Infer ESM2.")
    parser.add_argument(
        "--checkpoint-path",
        type=Path,
        required=True,
        help="Path to the ESM2 pretrained checkpoint",
    )
    parser.add_argument(
        "--data-path",
        type=Path,
        required=True,
        help="Path to the CSV file containing sequences and label columns",
    )
    parser.add_argument("--results-path", type=Path, required=True, help="Path to the results directory.")

    parser.add_argument(
        "--precision",
        type=str,
        choices=get_args(PrecisionTypes),
        required=False,
        default="bf16-mixed",
        help="Precision type to use for training.",
    )
    parser.add_argument(
        "--num-gpus",
        type=int,
        required=False,
        default=1,
        help="Number of GPUs to use for training. Default is 1.",
    )
    parser.add_argument(
        "--num-nodes",
        type=int,
        required=False,
        default=1,
        help="Number of nodes to use for training. Default is 1.",
    )
    parser.add_argument(
        "--micro-batch-size",
        type=int,
        required=False,
        default=2,
        help="Micro-batch size. Global batch size is inferred from this.",
    )
    parser.add_argument(
        "--pipeline-model-parallel-size",
        type=int,
        required=False,
        default=1,
        help="Pipeline model parallel size. Default is 1.",
    )
    parser.add_argument(
        "--tensor-model-parallel-size",
        type=int,
        required=False,
        default=1,
        help="Tensor model parallel size. Default is 1.",
    )
    parser.add_argument(
        "--prediction-interval",
        type=str,
        required=False,
        choices=get_args(IntervalT),
        default="epoch",
        help="Intervals to write DDP predictions into disk",
    )
    parser.add_argument(
        "--include-hiddens",
        action="store_true",
        default=False,
        help="Include hiddens in output of inference",
    )
    parser.add_argument(
        "--include-input-ids",
        action="store_true",
        default=False,
        help="Include input_ids in output of inference",
    )
    parser.add_argument(
        "--include-embeddings",
        action="store_true",
        default=False,
        help="Include embeddings in output of inference",
    )
    parser.add_argument(
        "--include-logits", action="store_true", default=False, help="Include per-token logits in output."
    )
    config_class_options: Dict[str, Type[BioBertConfig]] = SUPPORTED_CONFIGS

    def config_class_type(desc: str) -> Type[BioBertConfig]:
        try:
            return config_class_options[desc]
        except KeyError:
            raise argparse.ArgumentTypeError(
                f"Do not recognize key {desc}, valid options are: {config_class_options.keys()}"
            )

    parser.add_argument(
        "--config-class",
        type=config_class_type,
        default="ESM2Config",
        help="Model configs link model classes with losses, and handle model initialization (including from a prior "
        "checkpoint). This is how you can fine-tune a model. First train with one config class that points to one model "
        "class and loss, then implement and provide an alternative config class that points to a variant of that model "
        "and alternative loss. In the future this script should also provide similar support for picking different data "
        f"modules for fine-tuning with different data types. Choices: {config_class_options.keys()}",
    )
    return parser

infer_esm2_entrypoint()

在 geneformer 检查点和数据上运行推理的入口点。

源代码位于 bionemo/esm2/scripts/infer_esm2.py

def infer_esm2_entrypoint():
    """Entrypoint for running inference on a geneformer checkpoint and data."""
    # 1. get arguments
    parser = get_parser()
    args = parser.parse_args()
    # 2. Call infer with args
    infer_model(
        data_path=args.data_path,
        checkpoint_path=args.checkpoint_path,
        results_path=args.results_path,
        include_hiddens=args.include_hiddens,
        include_embeddings=args.include_embeddings,
        include_logits=args.include_logits,
        include_input_ids=args.include_input_ids,
        micro_batch_size=args.micro_batch_size,
        precision=args.precision,
        tensor_model_parallel_size=args.tensor_model_parallel_size,
        pipeline_model_parallel_size=args.pipeline_model_parallel_size,
        devices=args.num_gpus,
        num_nodes=args.num_nodes,
        config_class=args.config_class,
    )

infer_model(data_path, checkpoint_path, results_path, min_seq_length=1024, include_hiddens=False, include_embeddings=False, include_logits=False, include_input_ids=False, micro_batch_size=64, precision='bf16-mixed', tensor_model_parallel_size=1, pipeline_model_parallel_size=1, devices=1, num_nodes=1, prediction_interval='epoch', config_class=ESM2Config)

使用 PyTorch Lightning 在 BioNeMo ESM2 模型上运行推理。

参数

名称	类型	描述	默认值
data_path	路径	输入数据的路径。	必需
checkpoint_path	路径	模型检查点的路径。	必需
results_path	路径	保存推理结果的路径。	必需
min_seq_length	int	要填充的最小序列长度。这应至少等于数据集中最长序列的长度	1024
include_hiddens	bool	是否在输出中包含隐藏状态。默认为 False。	False
include_embeddings	bool	是否在输出中包含嵌入。默认为 False。	False
include_logits	(bool, 可选)	是否在输出中包含令牌 logits。默认为 False。	False
include_input_ids	(bool, 可选)	是否在输出中包含 input_ids。默认为 False。	False
micro_batch_size	int	用于推理的微批大小。默认为 64。	64
precision	PrecisionTypes	用于推理的精度类型。默认为 "bf16-mixed"。	'bf16-mixed'
tensor_model_parallel_size	int	用于分布式推理的张量模型并行大小。默认为 1。	1
pipeline_model_parallel_size	int	用于分布式推理的流水线模型并行大小。默认为 1。	1
devices	int	用于推理的设备数量。默认为 1。	1
num_nodes	int	用于分布式推理的节点数量。默认为 1。	1
prediction_interval	IntervalT	将 predict 方法输出写入磁盘以进行 DDP 推理的间隔。默认为 epoch。	'epoch'
config_class	Type[BioBertConfig]	用于使用提供的检查点配置模型的配置类	ESM2Config

源代码位于 bionemo/esm2/scripts/infer_esm2.py

def infer_model(
    data_path: Path,
    checkpoint_path: Path,
    results_path: Path,
    min_seq_length: int = 1024,
    include_hiddens: bool = False,
    include_embeddings: bool = False,
    include_logits: bool = False,
    include_input_ids: bool = False,
    micro_batch_size: int = 64,
    precision: PrecisionTypes = "bf16-mixed",
    tensor_model_parallel_size: int = 1,
    pipeline_model_parallel_size: int = 1,
    devices: int = 1,
    num_nodes: int = 1,
    prediction_interval: IntervalT = "epoch",
    config_class: Type[BioBertConfig] = ESM2Config,
) -> None:
    """Runs inference on a BioNeMo ESM2 model using PyTorch Lightning.

    Args:
        data_path (Path): Path to the input data.
        checkpoint_path (Path): Path to the model checkpoint.
        results_path (Path): Path to save the inference results.
        min_seq_length (int): minimum sequence length to be padded. This should be at least equal to the length of largest sequence in the dataset
        include_hiddens (bool, optional): Whether to include hidden states in the output. Defaults to False.
        include_embeddings (bool, optional): Whether to include embeddings in the output. Defaults to False.
        include_logits (bool, Optional): Whether to include token logits in the output. Defaults to False.
        include_input_ids (bool, Optional): Whether to include input_ids in the output. Defaults to False.
        micro_batch_size (int, optional): Micro batch size for inference. Defaults to 64.
        precision (PrecisionTypes, optional): Precision type for inference. Defaults to "bf16-mixed".
        tensor_model_parallel_size (int, optional): Tensor model parallel size for distributed inference. Defaults to 1.
        pipeline_model_parallel_size (int, optional): Pipeline model parallel size for distributed inference. Defaults to 1.
        devices (int, optional): Number of devices to use for inference. Defaults to 1.
        num_nodes (int, optional): Number of nodes to use for distributed inference. Defaults to 1.
        prediction_interval (IntervalT, optional): Intervals to write predict method output into disck for DDP inference. Defaults to epoch.
        config_class (Type[BioBertConfig]): The config class for configuring the model using checkpoint provided
    """
    # create the directory to save the inference results
    os.makedirs(results_path, exist_ok=True)

    # Setup the strategy and trainer
    global_batch_size = infer_global_batch_size(
        micro_batch_size=micro_batch_size,
        num_nodes=num_nodes,
        devices=devices,
        tensor_model_parallel_size=tensor_model_parallel_size,
        pipeline_model_parallel_size=pipeline_model_parallel_size,
    )

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

    prediction_writer = PredictionWriter(output_dir=results_path, write_interval=prediction_interval)

    trainer = nl.Trainer(
        accelerator="gpu",
        devices=devices,
        strategy=strategy,
        num_nodes=num_nodes,
        callbacks=[prediction_writer],
        plugins=nl.MegatronMixedPrecision(precision=precision),
    )

    dataset = InMemoryCSVDataset(data_path=data_path)
    datamodule = ESM2FineTuneDataModule(
        predict_dataset=dataset,
        micro_batch_size=micro_batch_size,
        global_batch_size=global_batch_size,
        min_seq_length=min_seq_length,
    )

    config = config_class(
        params_dtype=get_autocast_dtype(precision),
        pipeline_dtype=get_autocast_dtype(precision),
        autocast_dtype=get_autocast_dtype(precision),
        include_hiddens=include_hiddens,
        include_embeddings=include_embeddings,
        include_input_ids=include_input_ids,
        skip_logits=not include_logits,
        tensor_model_parallel_size=tensor_model_parallel_size,
        pipeline_model_parallel_size=pipeline_model_parallel_size,
        initial_ckpt_path=str(checkpoint_path),
        initial_ckpt_skip_keys_with_these_prefixes=[],  # load everything from the checkpoint.
    )

    tokenizer = get_tokenizer()
    module = biobert_lightning_module(config=config, tokenizer=tokenizer)

    # datamodule is responsible for transforming dataloaders by adding MegatronDataSampler. Alternatively, to
    # directly use dataloader in predict method, the data sampler should be included in MegatronStrategy
    trainer.predict(module, datamodule=datamodule)  # return_predictions=False failing due to a lightning bug