Scott Stevenson

Viewing Jupyter notebooks at the command line

The Jupyter notebook is a literate programming environment which has become ubiquitous in scientific computing. While the standard tools for interacting with notebooks are web applications, it’s often useful to be able to view notebooks at the command line. I find this convenient when logged into a remote machine via SSH, and the process of configuring SSH to forward a port, starting a Jupyter server, and navigating to it in a web browser is a chore just to view a notebook for a few seconds.

Viewing other literate programming formats such as R Markdown and Pweave at the command line is trivial: these use lightweight markup languages which don’t need special rendering to be legible. As such it’s sufficient to print the raw file contents to the terminal, for example with cat.

This isn’t the case for Jupyter notebooks. Internally, notebooks are structured JSON documents with a strict schema. Notebooks can contain images and other binary data, which if viewed with a pager will print unintelligible Base64 encoded data to your terminal. Characters such as " are backslash-escaped, making copying code from the output frustrating.

Instead of printing the notebook file directly to the terminal, we’ll convert it to another format first. nbconvert is a program to convert notebooks to rich formats including HTML and PDF, as well as plain text such as Markdown. By design, Markdown syntax is unobtrusive and readable in plain text without rendering, so it’s a good choice for our output format.

As an example, we’ll use a notebook which contains both prose and code cells taken from the documentation for JAX.

A Jupyter notebook in JupyterLab

We can use nbconvert to convert the notebook to Markdown, printing the output to standard output.

$ jupyter nbconvert --stdout --to markdown jax_quickstart.ipynb
[NbConvertApp] Converting notebook jax_quickstart.ipynb to markdown
# JAX Quickstart

**JAX is NumPy on the CPU, GPU, and TPU, with great automatic
differentiation for high-performance machine learning research.**

import jax.numpy as jnp
from jax import grad, jit, vmap
from jax import random

## Multiplying Matrices

We'll be generating random data in the following examples. One big
difference between NumPy and JAX is how you generate random numbers. For
more details, see [Common Gotchas in JAX].

[Common Gotchas in JAX]:

key = random.PRNGKey(0)
x = random.normal(key, (10,))

Let's dive right in and multiply two big matrices.

size = 3000
x = random.normal(key, (size, size), dtype=jnp.float32)
%timeit, x.T).block_until_ready()  # runs on the GPU

We added that `block_until_ready` because [JAX uses asynchronous
execution by default](

JAX NumPy functions work on regular NumPy arrays.

nbconvert writes the header [NbConvertApp] Converting notebook jax_quickstart.ipynb to markdown to standard error and the notebook body to standard output, so we can filter out the header by redirecting standard error.

jupyter nbconvert --stdout --to markdown jax_quickstart.ipynb 2>/dev/null

This redirection syntax is for Bourne-like shells such as Bash. If you’re using an alternative such as fish, change as appropriate.

This is already more readable than the raw JSON document, but we can improve the readability of the code cells by applying syntax highlighting. We’ll pipe the output of nbconvert to pygmentize, a command line interface to the Pygments syntax highlighting library. Pygments is a dependency of nbconvert, so if you’ve followed to here you’ll already have it installed.

As we’re piping text to pygmentize on standard input, there’s no filename from which to determine the language of the input so we specify it using the -l flag.

jupyter nbconvert --stdout --to markdown jax_quickstart.ipynb 2>/dev/null \
  | pygmentize -l md

We can pipe the output of pygmentize to a pager like less to scroll through and search within the notebook.

jupyter nbconvert --stdout --to markdown jax_quickstart.ipynb 2>/dev/null \
  | pygmentize -l md | less -R

The -R flag instructs less to output ANSI escape sequences in raw form in order to colour the output. If we run this command, we’ll get a legible rendering of the notebook with syntax highlighting. Thanks to less, we can scroll up and down through the notebook, and search within it by pressing / and typing a query.

nbless output in a terminal

However, this is a longer command than is convenient to type for each use, so let’s create a shell script to encapsulate it, and also take the opportunity to implement some input validation.

#!/usr/bin/env bash

set -euo pipefail

readonly EX_USAGE=64
readonly EX_NOINPUT=66
readonly EX_UNAVAILABLE=69

usage() {
  echo "usage: $(basename "$0") [-h] NOTEBOOK"
  echo "Convert a Jupyter notebook to Markdown and view with less"

if [[ $# -ne 1 ]]; then
  exit $EX_USAGE

if [[ "$1" = -h || "$1" = --help ]]; then

if [[ ! -f "$1" ]]; then
  echo "$(basename "$0"): $1: No such file"
  exit $EX_NOINPUT

if ! command -v jupyter >/dev/null; then
  >&2 echo "$(basename "$0"): command not found: jupyter"

jupyter nbconvert --stdout --to markdown "$1" 2>/dev/null | pygmentize -l md \
  | less -R

If we save this as an executable script on our shell’s command search path, for example as nbless, next time we’re at the command line on a remote machine and want to view a notebook it’s as simple as nbless notebook.ipynb—no port forwarding, Jupyter server, or web browser required.