test-myst-directives

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

MyST Directives Test Page

Created On: May 19, 2026 | Last Updated On: May 19, 2026

This page tests all common MyST and sphinx-design directives to verify they render correctly with the PyTorch Sphinx theme.

Dropdowns

Basic Dropdown

More details on mutation and aliasing

custom_op asks for a precise mutation and aliasing contract because PyTorch uses that contract in FakeTensor, autograd, functionalization, and torch.compile.

For a functional custom_op, PyTorch assumes the operator does not mutate any input and that returned tensors are fresh values. This is the easiest kind of op to set up for torch.compile.

For an in-place custom_op, torch.Tag.inplace gives PyTorch a stronger and more specific contract: the first argument and the returned object are the same object. This lets PyTorch derive the fake-tensor behavior from the schema instead of requiring a separate fake kernel.

Dropdown with Title and Open by Default

Click to expand this section

This dropdown is open by default. It contains some code:

import torch
x = torch.randn(3, 4)
print(x)

Dropdown with Admonition Style

Important implementation note

This is a dropdown styled as a warning. Be careful when using this API as it may change in future versions.

• Point one about the API

• Point two about compatibility

• Point three about deprecation timeline

Tip: Performance optimization

Use torch.compile for best performance:

model = torch.compile(model)

Admonitions

Note

This is a note admonition. Use it to highlight important information.

Warning

This is a warning. Proceed with caution when using experimental APIs.

Tip

Use torch.no_grad() context manager to disable gradient computation during inference for better performance.

Important

Always call model.eval() before inference and model.train() before training.

Caution

Mixing CPU and CUDA tensors in operations will raise a RuntimeError.

Danger

Never store sensitive data in model checkpoint files without encryption.

See also

See the PyTorch documentation for more details.

Hint

You can use torch.cuda.is_available() to check for GPU availability.

Custom Titled Admonition

Custom Title Here

This is an admonition with a custom title styled as a tip.

Tab Sets

Python

Use Python for most deep learning tasks:

import torch
x = torch.tensor([1.0, 2.0, 3.0])

C++

Use the C++ frontend for production:

#include <torch/torch.h>
auto x = torch::tensor({1.0, 2.0, 3.0});

CLI

Use the CLI for quick experiments:

python -m torch.utils.benchmark

Cards

Standalone Card

Getting Started with PyTorch

Learn the basics of PyTorch with our beginner tutorials. Covers tensors, autograd, neural networks, and more.

Footer content

https://pytorch.org/tutorials/

Card Grid

Training

Learn how to train models with PyTorch.

https://pytorch.org

Deployment

Deploy models to production with TorchServe.

https://pytorch.org

Badges and Buttons

plain badge primary secondary success warning danger info light dark

Outline Badges

primary outline warning outline success outline

Buttons

Go to PyTorch

View Documentation

Definition Lists

Term 1

Definition of the first term. This can span multiple lines.

Term 2

Definition of the second term.

torch.Tensor

The main data structure in PyTorch. A multi-dimensional matrix containing elements of a single data type.

Field Lists

Author:

PyTorch Team

Version:

2.0

Status:

Stable

License:

BSD-3

Task Lists

• Install PyTorch

• Set up development environment

• Train first model

• Deploy to production

Substitutions

This project is called PyTorch version 2.0.

Code Blocks

import torch
import torch.nn as nn

model = nn.Linear(10, 5)
x = torch.randn(3, 10)
output = model(x)
print(output.shape)

Installation command

pip install torch torchvision torchaudio

Math

\[\nabla_\theta J(\theta) = \mathbb{E}_{\pi_\theta} \left[ \sum_{t=0}^{T} \nabla_\theta \log \pi_\theta(a_t|s_t) \cdot G_t \right]\]

Inline math: \(E = mc^2\)

Tables

Comparison of Optimizers

Optimizer	Description	Learning Rate	Momentum
SGD	Stochastic Gradient Descent	0.01	0.9
Adam	Adaptive Moment Estimation	0.001	N/A
AdamW	Adam with weight decay	0.001	N/A

Figures

The PyTorch logo.

Block Attributes

The attrs_block extension lets you attach HTML attributes (classes, IDs) to the next block element using {.classname} or {#id} syntax.

This paragraph is styled with the sd-text-primary class via attrs_block.

This paragraph has a light background, padding, and rounded corners.

Toggles (Nested Dropdowns)

Outer dropdown

This is the outer content.

Inner dropdown

This is nested inside the outer dropdown.

Sphinx-Design Containers

Column 1

Content in the first column of a responsive grid.

Column 2

Content in the second column.

Column 3

Content in the third column. On mobile, all columns stack vertically.