🔒 Processed locally in your browser — no data uploaded
← Back to Home

Dataset Format Converter

Input Format
Output

What the Dataset Format Converter does

This converter rewrites a tabular dataset between JSONL, CSV, JSON array and TSV inside your browser, without uploading the data anywhere. You pick the format you are pasting in, choose the format you want out, and the converted text appears in the output panel below with a count of how many records and how many distinct fields it found in the pasted data.

Three JSONL lines such as {"name":"Alice","age":30,"role":"engineer"}, one for Bob and one for Charlie, turn into a CSV with a name,age,role header and three rows, or into the same three objects indented as a JSON array. Field names are collected across every record, so a key that appears in only one row still becomes a column.

How to use it

  1. Pick the input formatSet the first menu to JSONL for one object per line, CSV for comma-separated rows with a header line, or JSON array for a bracketed list of objects.
  2. Paste your dataDrop the text into the box below the menu. Conversion is triggered by changing a format menu, so paste first and then touch a selector to fire it.
  3. Choose the output formatSet the output menu to JSONL, CSV, JSON array or TSV. The preview panel, the record count and the distinct field count all refresh together as soon as that menu changes.
  4. Download the full resultThe preview is cut off after ten thousand characters, so use the download button for the whole result. It saves as converted.jsonl, converted.csv, converted.json or converted.tsv depending on the output format you picked.

Options and formats

Input: JSONLEvery non-blank line is parsed as its own object. A syntax error on any single line aborts the whole conversion and puts the parser message on screen instead of the data.
Input: CSVThe first row becomes the header and each later row becomes one record. Values that look numeric are stored as numbers, and everything else stays as plain text.
Output: JSON arrayWrites every record as one indented array, which is the shape a dataset usually needs when it is read as a single file rather than streamed line by line.
Output: TSVUses tabs between fields and adds no quoting at all. Tab-separated files open cleanly in spreadsheet software and sidestep the quoting rules that CSV needs.
Download and clearDownload saves the whole converted output with the extension matching the selected format. Clear wipes the input textarea, the preview and both counters in one action.

Things to watch out for

Quoted commas break CSV inputThe CSV reader splits every line on commas before it looks for quotes, so a field like "Smith, John" is cut into two columns and every column after it lands in the wrong place.
Empty cells become zeroAn empty CSV cell fails the numeric check and is stored as the number zero, and anything that does parse as a number is converted, so a zip code of 01234 loses its leading zero.
Nested objects stringify badlyA record containing an object or array converts into CSV as unhelpful text, and the result cannot be read back into the original structure. Keep nested data in JSONL or JSON.
Typing does not re-trigger itEditing the textarea alone leaves the previous output on screen. After pasting new data you must change one of the two format menus, even if only back to the value it already had.

Frequently asked questions

Will fields that only some records have survive?

Yes. Column names are gathered from every record, so a key present in a single object becomes a column for the whole file and the other records get an empty value there. The order follows first appearance across the dataset rather than alphabetical order.

How much data can I paste in?

There is no fixed cap, but the whole set is parsed in the browser tab and the preview stops after ten thousand characters. Large inputs can make the page sluggish, so download the converted file instead of reading it on screen.

Does it handle nested JSON for training data?

It carries nested objects through JSONL and JSON arrays untouched, which is what chat and instruction formats need. Flattening the same records into CSV or TSV turns the nested parts into text you cannot rebuild, so stay in a JSON format for those.

What if one JSONL line is malformed?

A single bad line throws during parsing, and the entire conversion fails with the parser's own message even when the other lines were fine. There is no skip-bad-lines option, so the offending line has to be corrected or removed first.

Which file extension does the download use?

The extension follows the output format you selected, so you get converted.csv, converted.tsv, converted.jsonl or converted.json. The file contents come from exactly the same conversion step that fills the preview, so what you see on screen and what lands on disk match.

Can it open an Excel workbook?

No, this page takes pasted text only. Export the sheet as CSV from your spreadsheet program first and paste that, or copy the cells straight out of the grid, which arrives comma-separated and works with the CSV input option without further editing.