Unit ID: DLPY-M01-U03 Estimated active time: 25-35 minutes

The runtime is part of the lesson

In deep learning, the environment matters.

A notebook can fail because of code. It can also fail because the package version is different, a GPU is unavailable, a model download is blocked, or a file path is wrong.

So we treat runtime setup as learning evidence, not admin work.

CPU-first path

The required course path uses CPU-first examples.

That means:

• small datasets;
• short training loops;
• no cloud account for core completion;
• no required GPU;
• no hidden large downloads; and
• visible package checks.

If you have a GPU later, you may use it for optional experiments. But you should first understand the CPU version. A faster wrong workflow is still wrong.

Reproducibility

Reproducibility means someone can understand and rerun your work well enough to check it.

For this course, reproducibility means:

• imports are visible;
• random seeds are visible;
• data source is recorded;
• train/validation/test split is recorded;
• package versions can be checked;
• notebook cells run from top to bottom;
• generated artifacts are small and named; and
• limitations are written.

It does not mean every number will match forever on every machine. Deep learning can vary across hardware, package versions, and numerical settings. We still reduce avoidable confusion.

Environment facts to record

In your setup notes, record:

• operating system;
• Python version;
• PyTorch version after installation;
• whether CUDA is available;
• whether you used CPU or GPU;
• course package date; and
• any runtime problem.

Current course setup status

The course source includes a runtime-check notebook:

courses/deep-learning-pytorch-foundations/notebooks/runtime-check-pytorch-cpu.ipynb

The notebook checks:

• PyTorch import;
• tensor creation;
• autograd;
• a tiny training loop;
• loss decrease; and
• saving/loading a small state_dict artifact.

If PyTorch is not installed, this notebook will not run yet. That is not a course failure. It is a setup step to complete before notebook QA.

Mini practice

Write one sentence for each:

1. Why CPU-first is a good default for this course.
2. Why a GPU does not prove a better model.
3. Why a notebook should show the package version.

Good answers should mention learning clarity, reproducibility, and evidence.

Mark complete