[Question] Guidance on training steps vs. dataset size for LoRA fine-tuning

[fyting]

Hi there,

First of all, thanks for the great work on this repository!

I have a question regarding the relationship between the dataset size and the number of training steps during LoRA fine-tuning. I noticed that in the provided example configs, steps is often set to fixed values like 1500 or 2000.

Is there a rule of thumb or empirical formula for adjusting the steps if my dataset size varies significantly? For example, should I aim for a specific number of epochs, or is the fixed step count generally sufficient regardless of the data scale?

Any advice or experience would be greatly appreciated. Thanks!