同步批归一化与分片状态
保持统计信息与权重一致
同步批归一化与分片状态 是 CoddyKit 上的免费 Deep Learning Academy 课时。 这是第 3 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 Deep Learning Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Deep Learning Academy 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
The Small-Batch Problem
Batch norm computes stats from each GPU's local batch. Split across many GPUs, each per-device batch shrinks and those stats get noisy.
Enter SyncBatchNorm
SyncBatchNorm fixes this by computing mean and variance across all GPUs together, as if it saw the full global batch.
Convert in One Call
You do not rewrite layers by hand. One helper converts every BatchNorm in your model to its synced version.
import torch.nn as nn
model = nn.SyncBatchNorm.convert_sync_batchnorm(model)Convert Before Wrapping
Order matters: call the conversion before you wrap the model in DDP, so the synced layers are the ones DDP manages.
model = nn.SyncBatchNorm.convert_sync_batchnorm(model)
model = DDP(model, device_ids=[local_rank])It Costs Communication
Syncing stats means an extra all-reduce at every batch norm layer. Use it when small per-GPU batches hurt accuracy, not by default.
The Memory Wall
Plain DDP copies the full model, gradients, and optimizer state onto every GPU. For huge models that redundancy wastes memory fast.
Shard the State
Sharding splits those tensors across GPUs so each device stores only a slice. Together the GPUs still hold the whole model.
Meet FSDP
PyTorch's FullyShardedDataParallel shards parameters, gradients, and optimizer state. It lets you train models far larger than one GPU's memory.
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
model = FSDP(model)Gather Just in Time
FSDP gathers each layer's full weights only when it is needed for compute, then frees them again. That keeps peak memory low.
ZeRO Stages
Sharding comes in levels, called ZeRO stages: shard optimizer state, then gradients, then parameters. More sharding saves more memory.
Pick the Right Tool
If your model fits per GPU, plain DDP is simplest. When it does not, reach for FSDP to shard state and keep going.
Quick Check
Decide what each technique is really for.
Recap
You met two consistency tools: SyncBatchNorm keeps stats global across GPUs, and FSDP shards state so giant models fit. Use each only when you need it.
常见问题解答
「同步批归一化与分片状态」课时是免费的吗？
是的 — 「同步批归一化与分片状态」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Deep Learning Academy 课程的其余内容，请升级到 CoddyKit PRO。 Deep Learning Academy 课程共包含 4 节课。
「同步批归一化与分片状态」这节课中我会学到什么？
保持统计信息与权重一致 你通过在浏览器中直接运行的动手代码来练习 Deep Learning Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 Deep Learning Academy 需要有经验吗？
无需任何先前经验。CoddyKit 上的 Deep Learning Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 3 节课，共 4 节。
「同步批归一化与分片状态」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 Deep Learning Academy 课中编写并运行代码吗？
能。每节 Deep Learning Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。