Absolute and Relative Error for LLM Training

[Elisa M.]

Hi everyone,

I'm currently working with Rulex software to train an LLM model, and I'm facing issues with overfitting. To address this, I'm fine-tuning some parameters, but I need some clarification.

Specifically, I'm trying to understand the difference between the "maximum absolute error" parameter that I can specify in the model and the option to use "relative error." How do these two parameters differ and interact?

Thank you in advance for your assistance!

Best regards,

Elisa

[Enrico]

Hi @Elisa M.

Have you already had the opportunity to take a look at this page on our documentation? https://doc.rulex.ai/docs/v12/factory/tasks/classification/trainlm.html

Let us know if you have any additional doubts