Dataset Editor: some entries are invalid

  • Resolved studio1337

    (@studio1337)


    6 months, 1 week ago

    I’m running into an issue I’m having a hard time resolving. I’m testing the plug-in on a (somewhat) throwaway site with very limited set of posts, to make sure I understand how everything runs before taking it to a larger site (25,000-30,000 articles and growing).

    I used the generator to create questions based on the contents of 13 posts, and I got this message at the top of the page: “Some entries are invalid, for example those ones:?1,?2,?3,?4,?5,?6,?7,?8,?9,?10, and?224?more.” There are 234 total, so it appears every entry is invalid.

    I checked this page for answers – https://meowapps.com/wordpress-chatbot-finetuned-model-ai/ – but the interface appears different from what I’m seeing. I don’t see “Format With Defaults,” and manually adding the “###” delimiter to one of the questions didn’t reduce the number of invalid entries. I also don’t see red Xs or check marks. I’m missing a step or doing something wrong, but I can’t figure out what that might be. I would be grateful for any assistance you can give to get past this hurdle.

Viewing 5 replies - 1 through 5 (of 5 total)
  • Plugin Support Val Meow

    (@valwa)

    Hey @studio1337! ??

    This documentation references an old version of the plugin, so many things have changed since then. An entry can be invalid for multiple reasons. Make sure you follow these rules:

    1. If the?messages?property of the entry is not present or if it contains less than 3 messages.
    2. If the roles of the first three messages are not in the order of ‘system’, ‘user’, ‘assistant’.
    3. If any two consecutive messages from the fourth message onwards have the same role.
    Thread Starter studio1337

    (@studio1337)

    Thank you for your response. I’m not sure how the instructions correspond with the UI (I see Question/Answer sets – 236 total), but I exported and opened the JSON file, and I see each message as its own entry in the array, and each one is set up just like this:

    {
    “messages”: [
    {
    “role”: “user”,
    “content”: “What is XXXXX?”
    },
    {
    “role”: “assistant”,
    “content”: “XXXXX is a (followed by definition).”
    }
    ]
    },

    There’s no system role showing in the file, and all messages are in the identical format above, the first three and all following. Did I miss a step in the setup somewhere? I’ve been using older documentation to get things set up apparently.

    knowitallnz

    (@knowitallnz)

    HI @studio1337 Studio, I have the exact issue. Were you able to resolve this please? And how? Mine also do not have “System” Do I just manually add “System” after role and leave it empty?

    IN case someone Else comes accross this issue. I received a reply from MEow support(they are awesome!)

    My AI Engine version with error was 2.5.0, updated to 2.5.2 resolved the issue

    See below the reply:

    Before using fine-tuning, you might be interested in reading this documentation. Using less than 3000 entries for fine-tuning won’t produce any reliable training results.

    The generator doesn’t create individual system prompts. You can activate Expert mode and manually add the “system” prompt by clicking the “+” button in the actions. We will address this issue in an upcoming version. Thank you for your patience and understanding.

    Hope this helps ??

    • This reply was modified 3 months, 4 weeks ago by knowitallnz.
    • This reply was modified 3 months, 4 weeks ago by knowitallnz.
    Plugin Author Jordy Meow

    (@tigroumeow)

    Hi @knowitallnz! The new version released yesterday fixes that ??

Viewing 5 replies - 1 through 5 (of 5 total)
  • You must be logged in to reply to this topic.

Tags