onrup

Data

ChatML

What is chatml?

ChatML is a conversation format that marks each message with an explicit role — system, user or assistant — using special tokens the model was trained to recognise. It is the trainer-ready target format most fine-tuning pipelines convert into.

Role markers let the model distinguish instruction from content from its own prior output. Without them a conversation is an undifferentiated block of text, and the model has to infer structure from wording alone.

The exact special tokens differ between model families. This is why a dataset formatted for one family can degrade output on another: the tokens that should be structural markers become ordinary text the model tries to reproduce.

Related terms

All terms in the glossary →

Start with the free tier

A magic link creates your account, your tenant and your first API key. No card until you ask for compute.