JSONL Token Estimator
Paste or drop a .jsonl file to check every line parses and see token counts for training or batch jobs.
Runs in your browser — nothing is uploadedSummary
—
| Line | Tokens | Preview |
|---|
How to use JSONL Token Estimator
- Paste JSONL or drop a .jsonl file.
- Invalid lines are listed with their line numbers.
- Read total, average and maximum tokens, and multiply by epochs for training.
Formula & assumptions
- Counts are estimates: the tokenizer approximation was calibrated against OpenAI’s o200k_base (typically within a few percent for English prose and code; up to about 20% off for some other languages). Nothing you paste leaves your browser.
Questions
What counts as a token here?
The text content of each line — for chat-format lines, the text of every message plus a small per-message overhead — estimated with the same approximation as the AI Token Counter.
How accurate is it?
Counts are estimates: the tokenizer approximation was calibrated against OpenAI’s o200k_base (typically within a few percent for English prose and code; up to about 20% off for some other languages). Nothing you paste leaves your browser.