Which formats does it check?

OpenAI chat messages, prompt and completion, Alpaca (instruction, input, output), ShareGPT (conversations with from and value) and preference pairs for DPO. The shape is detected per line, so a file that mixes them is reported line by line.

Is the split repeatable?

Yes. The train and validation split uses a seeded shuffle, so the same seed gives the same split every time.

Are my files uploaded?

No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.

Is it free? Are there limits or watermarks?

Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.

Does it work offline?

Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Home / AI Builder / Fine-tune Dataset Checker

Fine-tune Dataset Checker

Validate a JSONL fine-tuning file line by line: roles, empty replies, duplicates, token limits, then download clean train and validation splits.

Free, no uploads, no sign-up. Runs 100% in your browser.

How Fine-tune Dataset Checker works

Everything runs in your browser. Prompts, datasets and model files are never uploaded. Each line is parsed and its shape detected: chat messages, prompt and completion, Alpaca, ShareGPT or preference pairs (DPO). Rules check roles, order, empty content and weights, and repeated rows are found by comparing the conversation text. Token counts come from a quick estimate or a real tokenizer.

Input: JSONL, JSON. Output: JSONL.

How to use Fine-tune Dataset Checker

Paste JSONL or drop a .jsonl file.
Pick a tokenizer and set your token limit and epochs.
Press Check dataset and fix the listed lines.
Download the clean rows, or a seeded train and validation split.

Tips

Every chat row needs at least one assistant message to learn from.
Training cost is roughly tokens in the dataset times epochs times the price per million.

Fine-tune Dataset Checker FAQ

Which formats does it check?: OpenAI chat messages, prompt and completion, Alpaca (instruction, input, output), ShareGPT (conversations with from and value) and preference pairs for DPO. The shape is detected per line, so a file that mixes them is reported line by line.
Is the split repeatable?: Yes. The train and validation split uses a seeded shuffle, so the same seed gives the same split every time.
Are my files uploaded?: No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.
Is it free? Are there limits or watermarks?: Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.
Does it work offline?: Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Related tools

Dataset Format Converter: Convert fine-tuning data between OpenAI chat JSONL, ShareGPT, Alpaca, prompt and completion, and CSV, with an optional system prompt.
Token Counter: Count tokens with the real GPT-4o, GPT-4, Llama 3, Mistral, Qwen, Gemma or DeepSeek tokenizer, see each token, context use and cost.
JSON Formatter: Prettify, validate, minify and explore JSON as a tree.
MCP Config Builder: Build Model Context Protocol server configs for Claude Desktop, Cursor, VS Code, Windsurf, Claude Code and Gemini CLI, or convert one app's config to another.