I've been struggling with figuring out a good dataset for fine-tuning. Most of the ones that exist were purpose made for finetuning/training a model already.

Does anyone have any tips for creating sufficient datasets for finetuning specific workloads?