The problem with the temperature persists
I reported this bug to Venice support a while ago. The problem is that when setting the temperature ≥1 the text gets distorted. The support team recommended me to always have it at 0.8 but what's the point of always having it like that if I want my answers to be more random?


Log in to comment and vote
Comments1
Jack
Feb 27, 2025
If you want to try experimenting with higher temperature values you can lower top_p and you’ll see less gibberish. We generally recommend 0.8 for a balance between creative and structured output from the LLM. The ideal temperature can vary from model to model.