ChatML

One popular choice is the ChatML format, and this is a good, flexible choice for many use-cases. It looks like this:

{%- for message in messages %}
    {{- '<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n' }}
{%- endfor %}

Source: HF Chat Templates - see the article (HF docs) for full details.

OpenAI switching to ChatML was announced in Introducing ChatGPT and Whisper APIs where they write:

API: Traditionally, GPT models consume unstructured text, which is represented to the model as a sequence of “tokens.” ChatGPT models instead consume a sequence of messages together with metadata. (For the curious: under the hood, the input is still rendered to the model as a sequence of “tokens” for the model to consume; the raw format used by the model is a new format called Chat Markup Language⁠(opens in a new window) (“ChatML”).)

Code example (OpenAI; Python bindings)

import openai
 
completion = openai.ChatCompletion.create(
  model="gpt-3.5-turbo", 
  messages=[{"role": "user", "content": "What is the OpenAI mission?"}]
)
 
print(completion)

To learn more about the GPT-3.5 API, visit our Chat guide⁠(opens in a new window).

See also:

• The Introduction Of Chat Markup Language (ChatML) Is Important For A Number Of Reasons $\leftarrow$ quick explainer / exposition of ChatML format
• OpenAI Introduced Chat Markup Language (ChatML) Based Input To Non-Chat Modes by Cobus Greyling Medium
• chatml openai-python
• Chat Markup Language ChatML (Preview) - Azure