JSONL Validator & Viewer
Paste newline-delimited JSON to check every record. See the exact line and message for each broken row, preview valid rows as a table, and detect common fine-tuning dataset schemas.
How JSONL validation works
JSONL, sometimes written NDJSON, stores one complete JSON value per line. It is the default container for large language model fine-tuning datasets and streamed logs because a consumer can read a single record without buffering the whole file. This tool splits your input on newlines, skips blank lines, and runs the native browser parser on each remaining line. Every failure is reported with its 1-based line number and the raw engine message, so a trailing comma, a smart quote, or an unescaped newline is easy to locate. Valid object rows are collected into a table built from the union of their top-level keys, with long cells truncated for display only. The first row is inspected to label familiar shapes such as OpenAI chat, OpenAI legacy prompt and completion, and Anthropic style pairs. Nothing leaves the browser, so private training data stays local. For single large payloads, format them first with the JSON Formatter, then estimate upload size with the Token Counter.