Google Gemini Fine-Tune JSONL Validator

updated 11 May 2026 · Vertex AI supervised tuning format

Your training data never leaves this tab. Vertex AI uploads your file when a job starts; this pre-flight check is fully local.

Gemini's shape, in one line

{
  "systemInstruction": {"role": "system", "parts": [{"text": "You are helpful."}]},
  "contents": [
    {"role": "user",  "parts": [{"text": "Hi"}]},
    {"role": "model", "parts": [{"text": "Hello"}]}
  ]
}

Two things to remember:

• The assistant role is called model. Not assistant, not bot. Wrong role name is the #1 error this validator catches.
• Content lives in parts, not as a string. Even a plain text message is "parts": [{"text": "..."}], an array.

What the validator checks

• Each line is valid JSON.
• Top-level contents is present and is an array.
• Optional systemInstruction is an object with its own parts array.
• First entry role is user.
• Roles strictly alternate user / model.
• Every parts is a non-empty array; every part has either text, inlineData, or fileData.
• No assistant role anywhere (common copy-paste from OpenAI samples).

Common mistakes I catch

{"contents": [{"role": "assistant", "parts": [{"text": "broken"}]}]}
// Error: Gemini uses 'model', not 'assistant'.

{"contents": [{"role": "user", "parts": {"text": "Hi"}}]}
// Error: 'parts' must be an array.

{"contents": [{"role": "user", "content": "Hi"}]}
// Error: Gemini uses 'parts', not 'content'.

Multi-modal parts

Inline images use inlineData with a mime type and base64 data:

{"contents": [
  {"role": "user", "parts": [
    {"inlineData": {"mimeType": "image/jpeg", "data": "..."}},
    {"text": "What is this?"}
  ]},
  {"role": "model", "parts": [{"text": "A cat."}]}
]}

Frequently asked questions

Related tools

• OpenAI Fine-Tune ValidatorChat (messages) and legacy prompt/completion formats.
• Anthropic Fine-Tune ValidatorBedrock's top-level system + alternating messages.
• Llama Fine-Tune ValidatorLlama 3 chat and ShareGPT.
• Mistral Fine-Tune ValidatorMistral chat templates.
• Fine-tune dataset editorAuthor rows in a visual UI.