Skip to main content

Ai system error in longue texts form

Hi team! are you aware of a big issue with the current Llama for characters feature,theres often errors when you ask hard question that it cannot answer in fews words but long texts format theres a reccurring error where symbole of differents languages will be mixed up together and match up and create a long text of incorencies,rubish words,i don’t have took a specific exemple i should have to illustrate my point but it’s a error that only occurs when the ai need to formatate a huge load of texts to explains a topic.

Status: Rejected7 comments

Log in to comment and vote

Comments7

  • Jack

    Team•

    Apr 2, 2025

    If this issue comes back email [email protected]

  • Yellow Oak

    •

    Mar 16, 2025

    See example here:

    Check out my chat with Venice.ai: https://venice.ai/chat/e40b23fa-9f82-4633-bb23-ca5aad921fab?ref=jSy5sH#veniceShareKey=myCLduls59vYpgKYsqd871KDFD8%2BJ3GqEPNB3meFx04%3D&veniceShareNonce=dWT7DZkG2gRiD7MOXIKW4JxrWFB5RUUQ

    • Jack

      Team•

      Mar 16, 2025

      What’s your temperature and top_p setting on? If it’s >0.8 results like this will be more common.

      • Pink Tide

        •

        Mar 17, 2025

        I had top_p set to 1.0 and temp was at 1.1. Sounds like these are too high, I don’t remember setting these to those levels but guess I must have. I’ve lowered them. By the way another issue I’ve just found doing that - it’s impossible to set the temp to 0.8, it just jumps to 0.6 and won’t stick on 0.8. Every other increment ok but 0.8. Weird.

        • Jack

          Team•

          Mar 26, 2025

          Are you still seeing the issue with temperature settings not sticking? Any luck on reducing gibberish with a lower temp?

  • Yellow Oak

    •

    Mar 16, 2025

    I just got a similar issue today using the llama model, big load of text of random nonsense. It starts off ok but then soon starts with the random nonsense for a long time.

    • Jack

      Team•

      Mar 26, 2025

      Try lowering your temperature in your text generation settings