Thanks for providing the details and screenshot!
If you're setting max_tokens to 150-200 in JanitorAI, the model will try to generate responses up to that limit, but sometimes it might stop mid-sentence due to token limits, prompt length, or how JanitorAI handles the output. Here are some suggestions to reduce cut-off responses:
-
Lower the input prompt length:
If your messages and chat history are long, they take up tokens and leave less room for output. Try deleting or shortening the previous chat history if possible. -
Increase
max_tokensif possible:
Even if you prefer shorter replies, a setting of 200 tokens occasionally isn't enough for the model to finish its thought. Try increasing to 250β300 just to test if it helps; you can always trim the reply yourself. -
Try OpenAI Default settings:
Avoid strict limits or jailbreaking; use "OpenAI Default" generation settings as recommended for best results. -
Check for accidental prompt content:
Don't add custom instructions/jailbreaksβleave the extra prompt fields as a single space. -
JanitorAI limitation:
Sometimes responses get cut off from JanitorAI's side (site timeout, UI limit, etc.), not our API. If it happens even after adjusting token limits, consider trying a different website like VenusChat.AI, ChubAI, or SillyTavern to compare. -
Model/website may truncate: