For a ML problem, should parameters be tuned on samples or all dataset?

Deep Learning
1

I am trying to tune a lightgbm model, and not sure how to tune parameters appropriately. Looking forward to seek for advices and best practices. Thanks.