Two things here. • Can someone help answer the question about custom loss functions? • If this is something we should include in our documentation, I can file an issue.

Peter Schofield <pcschof@gmail.com>

On the front page of the documentation for timGPT, it says you can implement a custom loss function:

Custom loss function: Tailor the fine-tuning process with a custom loss function to meet specific performance metrics.

However the page it links you to (https://docs.nixtla.io/docs/tutorials-fine_tuning_with_a_specific_loss_function) only shows you how to use the standard loss functions (MAE, MSE, RMSE, SMAPE, and MAPE). Is there actually a way to implement a custom loss function?

Thanks!

To my knowledge, the user cannot specify a custom loss function. We can only choose from the list. Maybe we should rephrase that in the homepage, cause in the documentation, we say that we can select a function depending on the error metric, and we don't really mention a custom function.

Thanks, that would be good to update in the documentation then. @Marco could you create an issue or PR on that? Thank you!