Any suggestions to reduce the training epochs? like tuning lr or something else? #42
Description
I find training epochs are 15 epochs lavin. But for llava the epochs are only 3. I want to reduce the training epochs. Are there any suggestions on how to change the hyper-parameters to reduce the training epochs but maintain the similar results?