0Pricing
Deep Learning Academy · 课时

使用 autocast 与 GradScaler 进行混合精度训练

使用半精度数学运算，大幅提升速度

使用 autocast 与 GradScaler 进行混合精度训练 是 CoddyKit 上的免费 Deep Learning Academy 课时。 这是第 1 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。

What Mixed Precision Means

By default PyTorch does math in 32-bit floats. Mixed precision runs many operations in 16-bit instead, which is faster and uses far less memory.

Why Half Precision Is Faster

Modern GPUs have special tensor cores tuned for 16-bit math. Feeding them half-precision data can speed up training two or three times with little accuracy loss.

The Catch With 16-Bit

Half precision has a tiny range, so very small gradient values can round down to zero. That silent underflow stalls learning if you do nothing about it.

Meet autocast

Wrap your forward pass in autocast and PyTorch picks a safe precision per operation automatically. You never cast tensors by hand.

with torch.autocast(device_type='cuda'):
    out = model(x)

What autocast Wraps

Put only the forward pass and loss inside autocast. The backward call stays outside the block, where PyTorch handles precision for you.

with torch.autocast(device_type='cuda'):
    out = model(x)
    loss = loss_fn(out, y)

Meet GradScaler

A GradScaler fixes underflow by multiplying the loss by a large factor before backward, so small gradients stay big enough to survive in 16-bit.

scaler = torch.cuda.amp.GradScaler()

Scale the Loss

Instead of loss.backward(), call scaler.scale(loss).backward(). The scaler inflates the loss first so the resulting gradients do not vanish.

scaler.scale(loss).backward()

Step Through the Scaler

Run the optimizer with scaler.step, which quietly unscales the gradients back to normal size before applying the update.

scaler.step(optimizer)

Update the Scale Factor

Finish each step with scaler.update(). It grows the scale when things are stable and shrinks it if it ever spots an overflow.

scaler.update()

The Full AMP Step

Together these calls form one clean AMP iteration: zero grads, autocast forward, scaled backward, scaler step, then update.

optimizer.zero_grad()
with torch.autocast(device_type='cuda'):
    loss = loss_fn(model(x), y)
scaler.scale(loss).backward()
scaler.step(optimizer)
scaler.update()

When To Reach For It

Mixed precision shines on a recent NVIDIA GPU with large batches. On a plain CPU it gives little benefit, so save it for real training runs.

Quick Check

You enabled autocast but tiny gradients keep vanishing. What tool fixes that?

Recap

Wrap the forward in autocast for fast 16-bit math, then use a GradScaler to scale, step, and update so tiny gradients survive. ⚡

常见问题解答

「使用 autocast 与 GradScaler 进行混合精度训练」课时是免费的吗？

是的 — 「使用 autocast 与 GradScaler 进行混合精度训练」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。

「使用 autocast 与 GradScaler 进行混合精度训练」这节课中我会学到什么？

使用半精度数学运算，大幅提升速度 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 1 节课，共 4 节。

「使用 autocast 与 GradScaler 进行混合精度训练」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。

此课程中的所有课时

  1. 使用 autocast 与 GradScaler 进行混合精度训练
  2. 为大批次累积梯度
  3. 分析性能瓶颈
  4. 减少 GPU 内存使用
← 返回 Deep Learning Academy