使用分布式检查点 (DCP) 进行异步保存¶

作者： Lucas Pasqualin、Iris Zhang、Rodrigo Kumpera、Chien-Chin Huang

检查点通常是分布式训练工作负载关键路径中的瓶颈，随着模型和世界规模的增长，其成本越来越高。一种抵消此成本的优秀策略是并行、异步地进行检查点。下面，我们扩展了分布式检查点教程入门中的保存示例，以展示如何将此功能与 torch.distributed.checkpoint.async_save 轻松集成。

您将学到什么

如何使用 DCP 并行生成检查点
优化性能的有效策略

先决条件

PyTorch v2.4.0 或更高版本
分布式检查点教程入门

异步检查点概述¶

在开始使用异步检查点之前，了解其与同步检查点的区别和限制很重要。具体而言

内存需求 - 异步检查点通过首先将模型复制到内部 CPU 缓冲区来实现。
这很有帮助，因为它可以确保模型和优化器权重在模型仍在进行检查点时不会发生变化，但确实会将 CPU 内存提高 checkpoint_size_per_rank X number_of_ranks 倍。此外，用户应注意了解其系统的内存限制。具体来说，固定内存意味着使用 page-lock 内存，与 pageable 内存相比，这种内存可能很稀缺。
检查点管理 - 由于检查点是异步的，用户需要管理同时运行的检查点。通常情况下，用户可以
通过处理从 async_save 返回的 Future 对象来使用自己的管理策略。对于大多数用户，我们建议一次只限制一个异步请求的检查点，避免每个请求额外的内存压力。

import os

import torch
import torch.distributed as dist
import torch.distributed.checkpoint as dcp
import torch.multiprocessing as mp
import torch.nn as nn

from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from torch.distributed.checkpoint.state_dict import get_state_dict, set_state_dict
from torch.distributed.checkpoint.stateful import Stateful
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType

CHECKPOINT_DIR = "checkpoint"


class AppState(Stateful):
    """This is a useful wrapper for checkpointing the Application State. Since this object is compliant
    with the Stateful protocol, DCP will automatically call state_dict/load_stat_dict as needed in the
    dcp.save/load APIs.

    Note: We take advantage of this wrapper to hande calling distributed state dict methods on the model
    and optimizer.
    """

    def __init__(self, model, optimizer=None):
        self.model = model
        self.optimizer = optimizer

    def state_dict(self):
        # this line automatically manages FSDP FQN's, as well as sets the default state dict type to FSDP.SHARDED_STATE_DICT
        model_state_dict, optimizer_state_dict = get_state_dict(model, optimizer)
        return {
            "model": model_state_dict,
            "optim": optimizer_state_dict
        }

    def load_state_dict(self, state_dict):
        # sets our state dicts on the model and optimizer, now that we've loaded
        set_state_dict(
            self.model,
            self.optimizer,
            model_state_dict=state_dict["model"],
            optim_state_dict=state_dict["optim"]
        )

class ToyModel(nn.Module):
    def __init__(self):
        super(ToyModel, self).__init__()
        self.net1 = nn.Linear(16, 16)
        self.relu = nn.ReLU()
        self.net2 = nn.Linear(16, 8)

    def forward(self, x):
        return self.net2(self.relu(self.net1(x)))


def setup(rank, world_size):
    os.environ["MASTER_ADDR"] = "localhost"
    os.environ["MASTER_PORT"] = "12355 "

    # initialize the process group
    dist.init_process_group("nccl", rank=rank, world_size=world_size)
    torch.cuda.set_device(rank)


def cleanup():
    dist.destroy_process_group()


def run_fsdp_checkpoint_save_example(rank, world_size):
    print(f"Running basic FSDP checkpoint saving example on rank {rank}.")
    setup(rank, world_size)

    # create a model and move it to GPU with id rank
    model = ToyModel().to(rank)
    model = FSDP(model)

    loss_fn = nn.MSELoss()
    optimizer = torch.optim.Adam(model.parameters(), lr=0.1)

    checkpoint_future = None
    for step in range(10):
        optimizer.zero_grad()
        model(torch.rand(8, 16, device="cuda")).sum().backward()
        optimizer.step()

        # waits for checkpointing to finish if one exists, avoiding queuing more then one checkpoint request at a time
        if checkpoint_future is not None:
            checkpoint_future.result()

        state_dict = { "app": AppState(model, optimizer) }
        checkpoint_future = dcp.async_save(state_dict, checkpoint_id=f"{CHECKPOINT_DIR}_step{step}")

    cleanup()


if __name__ == "__main__":
    world_size = torch.cuda.device_count()
    print(f"Running async checkpoint example on {world_size} devices.")
    mp.spawn(
        run_fsdp_checkpoint_save_example,
        args=(world_size,),
        nprocs=world_size,
        join=True,
    )

使用固定内存获得更高性能¶

如果上述优化仍然不够高效，您可以利用针对 GPU 模型的额外优化，该优化使用固定内存缓冲区进行检查点暂存。具体来说，这种优化解决了异步检查点的主要开销，即将内存复制到检查点缓冲区。通过在检查点请求之间维护固定内存缓冲区，用户可以利用直接内存访问来加速此复制过程。

注意

此优化的主要缺点是在检查点步骤之间缓冲区的持久性。如果没有固定内存优化（如上所示），任何检查点缓冲区将在检查点完成后立即释放。使用固定内存实现，此缓冲区将在步骤之间维护，导致相同的峰值内存压力在应用程序生命周期内持续存在。

import os

import torch
import torch.distributed as dist
import torch.distributed.checkpoint as dcp
import torch.multiprocessing as mp
import torch.nn as nn

from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from torch.distributed.checkpoint.state_dict import get_state_dict, set_state_dict
from torch.distributed.checkpoint.stateful import Stateful
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
from torch.distributed.checkpoint import StorageWriter

CHECKPOINT_DIR = "checkpoint"


class AppState(Stateful):
    """This is a useful wrapper for checkpointing the Application State. Since this object is compliant
    with the Stateful protocol, DCP will automatically call state_dict/load_stat_dict as needed in the
    dcp.save/load APIs.

    Note: We take advantage of this wrapper to hande calling distributed state dict methods on the model
    and optimizer.
    """

    def __init__(self, model, optimizer=None):
        self.model = model
        self.optimizer = optimizer

    def state_dict(self):
        # this line automatically manages FSDP FQN's, as well as sets the default state dict type to FSDP.SHARDED_STATE_DICT
        model_state_dict, optimizer_state_dict = get_state_dict(model, optimizer)
        return {
            "model": model_state_dict,
            "optim": optimizer_state_dict
        }

    def load_state_dict(self, state_dict):
        # sets our state dicts on the model and optimizer, now that we've loaded
        set_state_dict(
            self.model,
            self.optimizer,
            model_state_dict=state_dict["model"],
            optim_state_dict=state_dict["optim"]
        )

class ToyModel(nn.Module):
    def __init__(self):
        super(ToyModel, self).__init__()
        self.net1 = nn.Linear(16, 16)
        self.relu = nn.ReLU()
        self.net2 = nn.Linear(16, 8)

    def forward(self, x):
        return self.net2(self.relu(self.net1(x)))


def setup(rank, world_size):
    os.environ["MASTER_ADDR"] = "localhost"
    os.environ["MASTER_PORT"] = "12355 "

    # initialize the process group
    dist.init_process_group("nccl", rank=rank, world_size=world_size)
    torch.cuda.set_device(rank)


def cleanup():
    dist.destroy_process_group()


def run_fsdp_checkpoint_save_example(rank, world_size):
    print(f"Running basic FSDP checkpoint saving example on rank {rank}.")
    setup(rank, world_size)

    # create a model and move it to GPU with id rank
    model = ToyModel().to(rank)
    model = FSDP(model)

    loss_fn = nn.MSELoss()
    optimizer = torch.optim.Adam(model.parameters(), lr=0.1)

    # The storage writer defines our 'staging' strategy, where staging is considered the process of copying
    # checkpoints to in-memory buffers. By setting `cached_state_dict=True`, we enable efficient memory copying
    # into a persistent buffer with pinned memory enabled.
    # Note: It's important that the writer persists in between checkpointing requests, since it maintains the
    # pinned memory buffer.
    writer = StorageWriter(cached_state_dict=True)
    checkpoint_future = None
    for step in range(10):
        optimizer.zero_grad()
        model(torch.rand(8, 16, device="cuda")).sum().backward()
        optimizer.step()

        state_dict = { "app": AppState(model, optimizer) }
        if checkpoint_future is not None:
            # waits for checkpointing to finish, avoiding queuing more then one checkpoint request at a time
            checkpoint_future.result()
        dcp.async_save(state_dict, storage_writer=writer, checkpoint_id=f"{CHECKPOINT_DIR}_step{step}")

    cleanup()


if __name__ == "__main__":
    world_size = torch.cuda.device_count()
    print(f"Running fsdp checkpoint example on {world_size} devices.")
    mp.spawn(
        run_fsdp_checkpoint_save_example,
        args=(world_size,),
        nprocs=world_size,
        join=True,
    )

结论¶

总之，我们学习了如何使用 DCP 的 async_save() API 从关键训练路径中生成检查点。我们还了解了使用此 API 引入的额外内存和并发开销，以及利用固定内存来进一步加速的额外优化。

使用分布式检查点 (DCP) 进行异步保存¶

异步检查点概述¶

使用固定内存获得更高性能¶

结论¶

文档

教程

资源