watsonx.ai

 View Only

  • 1.  LLM low-level detail questions

    Posted Sun December 01, 2024 02:39 PM

    When one trains a model, how does the training link in at the programmatic level?
    Modification of the model?
    Extension of the model?
    Storage of text?
    Etc.
    Trying to grasp the nitty-gritty!


    #watsonx.ai
    #GenerativeAI

    ------------------------------
    Jack Woehr
    Senior Consultant
    Seiden Group LLC
    Beulah CO
    3038478442
    ------------------------------


  • 2.  RE: LLM low-level detail questions

    Posted Mon December 02, 2024 10:04 AM

    I spent yesterday asking ChatGPT about all this .. continuing today .. very informative!



    ------------------------------
    Jack Woehr
    Senior Consultant
    Seiden Group LLC
    Beulah CO
    3038478442
    ------------------------------

    Original Message


  • 3.  RE: LLM low-level detail questions

    Posted Mon December 02, 2024 03:08 PM

    Hi Jack,

    When you train the LLM model, it embeds the new data into the existing model, and the model size increases. Currently, I don't see any mechanism to untrain the model on the previously trained data. So, whenever you train a model, it results in a bigger model size. 



    ------------------------------
    Kapil Rajyaguru
    ------------------------------

    Original Message