JSON Lines is the format your training data, your logs and your BigQuery exports arrive in, and almost nothing handles it natively. Validate and repair line by line, convert to CSV, SQL or Parquet, query with jq or SQL, split, sample, dedupe by near-duplicate similarity, and check a fine-tuning file against the OpenAI, Anthropic, Gemini or DPO schema before you pay to train on it.
What this kit does that a general converter doesn't.
The dataset tools. Chat-format conversion, token counting, leakage detection between train and eval, k-fold splitting — built for people preparing LLM training data.
Pick a tool
Grouped the way jsonlkit.com groups them. Every link opens the tool on its own site.
Convert
Inspect
Transform
- Remove Duplicate JSON / JSONL Records
- Split or Merge a JSONL / NDJSON File
- Merge JSONL Files
- Filter & View a JSONL File (No-Code, No jq)
- jq Playground for JSONL & NDJSON
- JSONL SQL Playground
- Search & Replace in JSON / JSONL (Regex)
- Sort JSONL Records by Any Key Path
- Shuffle, Sample, or Split a JSONL Training Set
- Flatten / Unflatten JSON & JSONL
- JSONL Anonymizer
- JSONL Join
- JSONL Shuffle
- Truncate JSONL
- JSONL Key Case Converter
- Gzip (.jsonl.gz)
- Keep or Drop JSONL Fields
- Rename JSONL Keys
- Remove Empty Fields from JSONL
- Split JSONL by Field Value
- Normalize Timestamps in JSONL
- Render JSONL Through a Template
- Chat Template Preview
- JSONL Generator
AI fine-tuning
- OpenAI Fine-Tune JSONL Validator
- Tool-Call JSONL Validator
- Claude (Anthropic) Fine-Tune JSONL Validator
- Gemini Fine-Tune JSONL Validator
- Llama Fine-Tune JSONL Validator
- Mistral Fine-Tune JSONL Validator
- Dataset editor
- Split JSONL into Train, Validation, Test
- JSONL K-Fold Split
- Train / Eval Leakage Detector
- DPO Dataset Validator
- OpenAI Batch API Validator
- Build Batch API Requests from JSONL
- JSONL Quality Filter
- Deduplicate Fine-Tune Prompts in JSONL
- Count Tokens in a JSONL Dataset (GPT, Claude, Llama)
- Fit JSONL Conversations Into a Token Budget
- Repair JSONL Conversations
Command line
Reading, not tools
Format explainers and sample files published on jsonlkit.com.
- What is JSONL? JSONL (JSON Lines) is a line-delimited JSON format — one JSON value per line, no enclosing array. Explained with examples, histor…
- JSONL Specification The full JSONL (JSON Lines) specification: line structure, encoding, line endings, MIME types, whitespace, escape rules, and the …
- JSONL vs JSON vs NDJSON vs CSV vs Parquet Compare JSONL (JSON Lines) with regular JSON, NDJSON, CSV, and Parquet. Side-by-side tables, conversion paths, and a decision gui…
- What is NDJSON? NDJSON (Newline-Delimited JSON) is a streaming JSON format — one JSON value per line, separated by . Explained with examples, MI…
- NDJSON vs JSONL NDJSON vs JSONL: side-by-side comparison of file extensions, MIME types, ecosystems, history, and edge cases. They are the same f…
- JSONL Examples Annotated JSONL examples: application logs, LLM fine-tuning datasets (OpenAI, Anthropic, Gemini), Elasticsearch _bulk, BigQuery i…
- JSONL Best Practices JSONL best practices: handling large files, streaming, compression, encoding, error recovery, schema evolution, libraries for Pyt…
- JSONL Troubleshooting Diagnose JSONL errors: Extra data, Unexpected token, Expecting value, trailing commas, BOM, CRLF, unescaped newlines, truncated f…
- JSONL FAQ Straight answers about JSON Lines: the extension to use, the MIME type, whether it needs a trailing newline, comments, arrays, st…
- JSONL Glossary Plain-English definitions of the terms around JSON Lines: NDJSON, record, streaming, backpressure, schema drift, flattening, cano…
- JSONL in Python Read and write JSON Lines in Python: the stdlib loop, pandas read_json(lines=True), Polars, orjson for speed, and streaming a hug…
- JSONL in JavaScript Parse JSON Lines in JavaScript: readline streaming in Node, async iterators, the browser Streams API for huge files, and writing …
- JSONL in Go Parse JSON Lines in Go with bufio.Scanner and json.Decoder, handle long lines, stream gzip, use json.RawMessage for mixed shapes,…
- JSONL on the Command Line Practical shell recipes for JSON Lines: count, filter and reshape with jq, split large files, sample with shuf, deduplicate, join…
- JSONL for Machine Learning How to prepare a JSONL training dataset: format per provider, deduplication, PII removal, train/val/test splitting, leakage check…
- JSONL for Structured Logging Why structured logs use JSON Lines: which fields to standardise, how to keep logs queryable, rotation and compression, and how to…
- Loading JSONL into BigQuery Load newline-delimited JSON into BigQuery: bq load and console steps, autodetect versus explicit schema, nested and repeated fiel…
- OpenAI Fine-Tuning JSONL Format The exact JSONL format OpenAI fine-tuning expects: messages array, role values, system prompts, weight, tool calls, and the rules…
- The ShareGPT Format The ShareGPT JSONL format used by Axolotl and LLaMA-Factory: the conversations array, from and value fields, role name variants, …