Tool
Flatten JSONL to CSV or TSV
Each JSON object becomes one table row. Use top-level keys as columns, or Alpaca and the first system/user/assistant turns. Nested values are JSON strings. Nothing leaves this tab.
Flattening runs in this tab. The JSONL is not uploaded. The download is built here, with a UTF-8 BOM so spreadsheet apps open Hangul.
How columns are chosen
Top-level keys: the union of object keys, in first-seen order, at most 80. Nested objects and arrays become JSON.stringify. Missing keys are empty cells.
Alpaca: instruction, input, output. Lines without string instruction and output are errors, not silent drops.
Chat: n_turns, first system, first user, first assistant, and messages_json. OpenAI messages, ShareGPT conversations, or Alpaca instruction/output can fill those turns. Extra turns stay only in messages_json.
CSV quotes fields that contain the delimiter, quotes, or newlines, doubling quotes. TSV uses a tab and leaves commas unquoted.
What this page does not do
- It does not guess a schema beyond top-level keys. Column types are not inferred.
- Excel formulas are not evaluated. Cells that start with = are still text.
- EUC-KR bytes are not decoded here. Decode on the CSV→JSONL page if the source table was CP949.
- At most 20,000 lines or about two million characters.