分析性能瓶颈
找出时间与内存的消耗位置
分析性能瓶颈 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。
Why Profile First
Before optimizing, find out where time actually goes. Guessing wastes effort, while a quick profile shows you the real slow spots.
Two Common Bottlenecks
Training usually stalls in one of two places: the GPU compute doing math, or the data pipeline feeding it. Knowing which one matters.
Time It Crudely First
Start simple by timing a loop section with the clock. A rough perf_counter reading often points you to the right area in seconds.
import time
t = time.perf_counter()
# run one batch
print(time.perf_counter() - t)GPU Work Is Async
CUDA runs in the background, so naive timers lie. Call synchronize first to make sure the GPU has truly finished before you read the clock.
torch.cuda.synchronize()The Built-In Profiler
For real detail, use the torch.profiler context manager. It records how long every operation takes on both CPU and GPU.
from torch.profiler import profileWrap the Code to Profile
Run the part you care about inside a profile block. Choosing both CPU and CUDA activities captures the whole picture.
with profile(activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA]) as prof:
model(x)Read the Table
Print results sorted by cost to see the heaviest ops at the top. The key_averages table groups identical operations together.
print(prof.key_averages().table(sort_by='cuda_time_total'))Spot a Data Bottleneck
If the GPU often sits idle waiting, your DataLoader is too slow. More workers or cached data usually fixes that gap.
Spot a Compute Bottleneck
If one matmul or conv dominates the table, the limit is raw compute. Mixed precision or a smaller model is the lever to pull.
Watch Memory Too
The profiler can also report peak memory. Tracking profile_memory reveals which layers eat the most, guiding what to trim.
with profile(profile_memory=True) as prof:
model(x)Measure, Change, Re-Measure
Optimization is a loop: profile, make one change, then profile again. Trust numbers, not hunches, to confirm a fix actually helped.
Quick Check
Your GPU often sits idle between batches. What is the likely bottleneck?
Recap
Profile before you tune: synchronize for honest timings, use torch.profiler to find the heaviest ops, then fix data or compute and measure again. 🔍
常见问题解答
「分析性能瓶颈」课时是免费的吗？
是的 — 「分析性能瓶颈」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。
「分析性能瓶颈」这节课中我会学到什么？
找出时间与内存的消耗位置 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。
「分析性能瓶颈」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。