[whimsicalism]

the linked resources are good, also the llava repo has pretty good guides on how to train which you can adapt to SFT