为大批次累积梯度
在较小的 GPU 上模拟大批次
为大批次累积梯度 是 CoddyKit 上的免费 Deep Learning Academy 课时。 这是第 2 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 Big-Batch Problem
Large batches often train more smoothly, but they also need lots of GPU memory. A small card simply cannot hold a giant batch at once.
The Core Trick
Gradient accumulation splits one big batch into small chunks. You add up their gradients and update once, as if the whole batch ran together.
Gradients Already Accumulate
PyTorch adds each backward pass into .grad rather than replacing it. This default behavior is exactly what accumulation relies on.
Pick an Accumulation Count
Choose how many mini-batches make one update. With accum_steps of four, four small batches behave like one batch four times larger.
accum_steps = 4Do Not Zero Every Step
The key change is timing: zero_grad only at the start of an accumulation cycle, not after every single mini-batch.
if step % accum_steps == 0:
optimizer.zero_grad()Scale the Loss
Divide each mini-batch loss by accum_steps before backward. This keeps the average gradient identical to running the full batch at once.
loss = loss_fn(model(x), y) / accum_steps
loss.backward()Step Only When Full
After enough mini-batches pile up, call optimizer.step. The accumulated gradients now reflect the whole large batch.
if (step + 1) % accum_steps == 0:
optimizer.step()The Full Pattern
Put it together: scale the loss, backward every step, but only step and zero once per cycle. The loop stays simple.
for step, (x, y) in enumerate(loader):
loss = loss_fn(model(x), y) / accum_steps
loss.backward()
if (step + 1) % accum_steps == 0:
optimizer.step()
optimizer.zero_grad()Memory Stays Small
You only ever hold one mini-batch in memory at a time. That is why a modest GPU can mimic a batch many times its real capacity.
The Trade-Off
Accumulation trades time for memory: more forward and backward passes per update mean each effective batch takes a little longer to finish.
Mind Batch Norm
Batch norm still sees only the small mini-batch, so its statistics are noisier than a true large batch would produce. Keep that in mind.
Quick Check
You accumulate over 4 mini-batches. When should you call optimizer.step()?
Recap
Split a big batch into chunks, divide the loss by accum_steps, backward every chunk, and step only once per cycle to fake a large batch on small memory. 🧮
常见问题解答
「为大批次累积梯度」课时是免费的吗？
是的 — 「为大批次累积梯度」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。
「为大批次累积梯度」这节课中我会学到什么？
在较小的 GPU 上模拟大批次 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 2 节课，共 4 节。
「为大批次累积梯度」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。