How to Finetune GPT-Like Large Language Models on a Custom Dataset