pytorch/tutorials

PyTorch tutorials.

What it is

A collection of PyTorch learning materials, published as web documentation at https://pytorch.org/tutorials. The tutorials are written as Python files that get converted into HTML pages, with a Jupyter notebook version automatically generated for each one (runnable in Google Colab).

Where to ask questions

If you have a question about a tutorial, post it on the dev-discuss forum — not as a GitHub issue. The README says you'll get a faster answer there.

How to report issues

Two types of issues are accepted:

  • Feature request — propose a new tutorial. You must explain why it's needed and how it shows PyTorch's value.
  • Bug report — report a broken or outdated tutorial. Include the output of python3 -m torch.utils.collect_env (which gathers info about your system) in the report.

How to contribute a new tutorial

  1. Write a Python file. If you want it executed when inserted into the docs, name it with the suffix tutorial (e.g., your_tutorial.py).
  2. Place it in one of these folders based on difficulty:
    • beginner_source — beginner level
    • intermediate_source — intermediate level
    • advanced_source — advanced level
    • recipes_source — short recipes
    • prototype_source — unstable prototype features
  3. For non-prototype tutorials, add it to the table of contents (toctree) and create a customcarditem in index.rst.
  4. For non-prototype tutorials, add a thumbnail in index.rst; for recipes, add one in recipes_index.rst.

If you start with a Jupyter notebook, there's a linked script to convert it to a Python file. After conversion, check that headings and structure are logical.

Building locally

The full build is large and needs a GPU. Options:

  • Full build (make docs): downloads data, runs the tutorials, builds docs to docs/. Takes 60–120 minutes on GPU systems.
  • Quick preview (make html-noplot): builds basic HTML to _build/html without running the tutorial code — useful if you don't have a GPU.

First install dependencies with pip install -r requirements.txt. You can use conda or virtualenv (the README shows a virtualenv setup).

Building a single tutorial

Set the GALLERY_PATTERN environment variable to a filename (or regex) to build only that tutorial. Example:

GALLERY_PATTERN="neural_style_transfer_tutorial.py" make html

Spell checking

Use pyspelling to catch spelling errors:

  • pyspelling — full check (~3 minutes)
  • pyspelling -n python — Python files only
  • pyspelling -n reST — .rst files only (currently limited to the beginner/ directory)

License

BSD licensed (see the LICENSE file in the repo).

Related

  • Project
    dotnet/machinelearning

    ML.NET is an open‑source, cross‑platform machine‑learning framework for .NET that provides data loading, transformation, a catalog of algorithms, and support for TensorFlow/ONNX models, all consumable via the `Microsoft.ML` NuGet package.

  • Project
    evilsocket/pwnagotchi

    An A2C-based reinforcement learning agent that learns to optimize the capture of crackable WPA WiFi handshakes and key material from its environment.

  • Project
    oumi-ai/oumi

    Oumi is an open‑source platform that covers the entire lifecycle of foundation models—data prep, training (SFT, LoRA, DPO, GRPO), tool‑using agentic fine‑tuning, multimodal support, evaluation with LLM judges, and fast inference/deployment via vLLM or cloud endpoints. It offers a CLI, Docker image, and a large library of ready‑to‑run YAML recipes for models ranging from 10 M to 400 B parameters.

  • Project
    unit8co/darts

    Darts is a Python library that simplifies time‑series forecasting and anomaly detection. It offers a unified `fit`/`predict` API for classic models (ARIMA, Exponential Smoothing) and deep‑learning models (N‑BEATS, TCN, Transformers), supports multivariate series, covariates, probabilistic forecasts, back‑testing, and integrates with PyTorch Lightning and MLflow.