Example for find tunning

  • Resolved daniz91

    (@daniz91)


    1 year, 10 months ago

    Hi, First thank you for creating this amazing plugin.

    I am trying to use it for education. Where student can ask questions in lessons, which works great, but the problem is that it gives untrue statements for the website.

    Example my site xyz.com paid membership cost $20

    If a student ask “what is the membership cost” it returns random answers such as. “The cost of our Membership plan is $9.99 per month. With a Premium Membership you will get access to exclusive content, special offers, and discounts. You also get access to our private community forum, exclusive discounts on partner products, and more.”

    In the new update fine tunning can you give me an example of how I can train or write it to have it show correct values or names? I want to use your add entry which seems simple for non-techy person like me.

    Thank you and appreciate this plugin

Viewing 4 replies - 1 through 4 (of 4 total)

  • This would be great. There is no explanation at all on how to fine tune it and prevent that it gives bad answers.

    Would be great if you can add this and be it with a Pro – Account ??

    Plugin Author Jordy Meow

    (@tigroumeow)

    Hi guys,

    Yes, it’s possible to train an AI model with the current version of the plugin. However, it is actually quite complex, and normally… this should done by an expert! ??

    The chatbot also needs to be aware (technically-speaking) of how you prepared the data, and how you did train the AI with it in details. There is an infinite number of ways of training an AI, and an infinite way to developer a chatbot too (even though they might all look similar). I am working now to find a simple process so that anyone can start training its own AI model and use it on the chatbot. And it will work ??

    Meanwhile, you can already start the biggest part of the work on your side: preparing the dataset! ?? Create two columns in a spreadsheet called “Prompt” and “Completion”, write a question in “Prompt”, and write the answer in “Completion”. You’ll need at least 500 rows, but it might be better to get over 1,000.

    It’s actually a bit more complex than this but the plugin will do the difficult task for you. You only need to create the content ??

    Do you you know about examples? Would you e.g. have questions and give the correct answer or ist it more to explaint the context of the individual implementation?

    Plugin Author Jordy Meow

    (@tigroumeow)

    In the case of a chatbot, the database would be all about questions and answers.

    Here is an example: https://gist.github.com/jordymeow/131f34e79bc9ee5e0763e34a29f9fc61. It’s a CSV file, so you can open it in Google Sheet, and then, export it from Google Sheet in CSV. That file can be uploaded directly in the “Create Dataset” module in AI Engine ?? You can also edit in there directly (it’s simple, and make sure you are formatting it correctly).

    This file is very small, too small to actually train a model properly. That said, I used it to train a model, and indeed, it kind of works. A train chatbot needs a lot, LOT of examples. To have good results, you would need around 1,000 – and preferably much more than that.

    I am currently trying to make a much bigger file (based on the one I shared), to see the effects. Then, I will be able to have better recommendations.

    Now, that said, I am also going to release a feature that will automatically generate those questions/answers based on your website, by reading it ?? That will be part of a Pro Version, however.

Viewing 4 replies - 1 through 4 (of 4 total)
  • The topic ‘Example for find tunning’ is closed to new replies.