IPYNB to PDF

An IPYNB file is a Jupyter Notebook: a JSON document that stores an entire working session as an ordered list of cells. Each cell is one of three types — markdown cells for narrative text, headings, tables and LaTeX math; code cells holding executable source plus the outputs from the last run; and raw cells that pass through untouched, often YAML front matter for a publishing pipeline. Code cells also carry their execution count, which is what Jupyter shows as In [3] and Out[3]. Because .ipynb is JSON, GitHub can render it but almost nothing else can, and it is a poor format for sharing with anyone who just needs to read the analysis. Converting to PDF fixes that. This converter reads all three cell types automatically, so you never have to think about which is which.

The built-in export path (File → Save and Export Notebook As → PDF, or jupyter nbconvert --to pdf) routes your notebook through LaTeX. If TeX Live, MiKTeX or Pandoc is not installed you get the notorious "500: Internal Server Error" or a nbconvert failed: xelatex not found message, and installing a full TeX distribution to export one document is a lot of gigabytes. Newer nbconvert offers --to webpdf, which renders through headless Chromium instead, but that means another install and a browser download. This tool takes the same shortcut without any setup: it parses the notebook JSON directly and writes the PDF with a PDF engine, in your browser. No LaTeX, no Chromium, no pip install, no template to pick.

A notebook PDF is only useful if it keeps the parts you actually ran. Markdown cells keep their heading hierarchy, bold and italic text, links, nested lists, blockquotes, horizontal rules and GFM tables. Code cells keep their indentation and get language-aware syntax highlighting, wrapped at the page width so nothing is clipped. Outputs are handled by type: stdout streams render as plain monospace, stderr gets a warning tint, execution results are labelled Out[n], and error tracebacks appear in a red block with the exception name and message on the first line. Charts and images — PNG, JPEG and SVG — are embedded at their original resolution. pandas DataFrames arrive as text/html and are rebuilt as genuine PDF tables with borders and a header row, rather than being flattened into text the way most converters do. LaTeX in markdown cells is typeset with KaTeX, and Mermaid diagrams are drawn too.

nbconvert is the right tool inside a scripted pipeline: it is reproducible, it takes templates, and it can run in CI. It is the wrong tool when you need one PDF now, on a machine where LaTeX is not installed, or when you hit a template mismatch after upgrading Jupyter. Common complaints — jupyter pdf missing output, images dropped from the export, the classic template no longer matching your version, an export that works locally but not on a colleague's laptop — all come from the LaTeX layer rather than from your notebook. An online converter has no LaTeX layer to go wrong. The trade-off is that it cannot execute your notebook: whatever outputs are saved in the file are what appear in the PDF. Run all your cells and save before exporting, and the results match.

Google Colab has no direct PDF export — the usual advice is to print the browser page, which loses long outputs and adds browser headers. Instead choose File → Download → Download .ipynb and convert that file here; you get proper pagination and searchable text. Notebooks written in the VS Code Jupyter extension are ordinary .ipynb files and work the same way, as do notebooks downloaded from a GitHub repository (use the Raw view, or download the repo, so you get the JSON rather than GitHub's rendered HTML page). Notebooks from Kaggle, Databricks, Deepnote and JupyterLite all export .ipynb too. Older files from the IPython era, which store cells under a worksheets key, are handled as well — most converters reject them outright.

Restart the kernel and run all cells before you export, so the saved outputs match the code and no cell shows a stale result. Use landscape orientation, or A3, when your notebook has wide DataFrames or charts; wide tables read far better across the page. Structure the notebook with markdown headings — they become the PDF's bookmark outline, which is what makes a long analysis navigable. Turn off the In[ ]/Out[ ] prompts for a report that should read like a document rather than a notebook, and turn off code cells entirely when you only want the narrative and the charts. Keep the light code theme for anything you plan to print. Finally, if a cell produced an interactive ipywidgets, Plotly or Bokeh view, re-run it with a static renderer — a live JavaScript widget has nothing to draw in a static PDF.