Questions about Training

Thanks for the support on my first year on Civitai, in January I trained my first Lora, though I have some questions about the training values, for previous attempts with a dataset of 40 images I set the repeat number to 8 on 17 epochs, and left the LR untouched. Currently I have 70 images for each Lora, should I lower the repeats to 4, set the epochs to 20 and increase the LR ? also should I slightly increase the Network Dim beyond 32 to store more info? For last I stick with the old Kohya engine and avoid the Beta new engine due to failed attempts. Thanks in advance