[Feature Request] "Generate only one line per request" Checkbox

[alicat22]

As a user, I'd like to be able to generate responses one paragraph at a time. Essentially this would just add a "\n" stop string to the API.

When using LLMs for story and adventure, having 1 paragraph generate at a time is incredibly helpful. It helps stop the story/adventure from progressing too far. Some models also have bad habits when they're allowed to reach the {eos}. For example, models trained on OAI logs will follow a pattern where the 3rd paragraph is a conclusion. This bias is detrimental to story/adventures.

Possible Solutions

• Generate only one line per request is how SillyTavern labels this checkbox.
• Multiline Replies is how Kobold Lite labels this checkbox.

Additional context
When using the OAI-compatible API, adding "\n" as a stop string does the following:
stop: [ "\\n" ]

Unfortunately, because an escape character is added, this option currently does not work.

This checkbox is also very user-friendly, as many frontends have adopted this option. Most users won't know what \n is.

If you require any additional details, please let me know!

[alicat22]

The OAI-compatible API endpoint I'm using is fixing the \ escape on \n, so this should allow \n to work properly.

I think leaving up this request still holds value, however, as many users won't know what \n is or does.