JSONL Validator

Check a fine-tuning dataset line by line.

How to use JSONL Validator

  1. Add your JSONL file. Drop a .jsonl file onto the uploader or pick it from your device. Only one file is accepted at a time.

  2. Check the summary. Read Total lines, Valid, Invalid and Format detected.

  3. Fix parse errors. Under Parse errors, each entry lists a line number and the problem, such as invalid JSON or a line that is not a JSON object.

  4. Review schema warnings. Under Schema warnings, look for records whose keys differ from the majority shape and, for chat data, problems with messages, roles or content.

Limits and privacy

  • The maximum file size is 25 MB.
  • The uploader accepts dataset files with .jsonl, .csv, .parquet, .arrow or .tsv extensions, but every line is parsed as JSON, so use it for JSONL.
  • Blank lines are skipped and do not count toward totals.
  • Each non-blank line must be a JSON object. Arrays, strings and numbers on a line are reported as errors.
  • Schema drift compares each record's sorted top-level keys with the most common set of keys in the file.
  • A file is treated as OpenAI chat format when more than half its valid records have a messages array. Chat checks then flag empty messages arrays, non-object messages, a missing role, roles other than system, user, assistant, tool or function, and non-string content.
  • The tool does not check token lengths or anything provider-specific beyond those message checks.
  • Validation runs in your browser. The file is not uploaded.

Frequently asked questions

How do I validate a fine-tuning JSONL file?

Upload it and check for parse errors and schema warnings. Each is listed with its line number so you can fix it in your editor.

What counts as a schema warning?

A record whose top-level keys differ from the most common key set in the file, or a chat-format record with an empty messages array, an unknown role or non-string content.

Which chat roles are accepted?

system, user, assistant, tool and function. Any other role value is flagged as not recognized.

Does it work on large files?

Files up to 25 MB are accepted. The whole file is read into memory in your browser, so very large inputs may be slow.

What's next