Finetune model on context without using prompt and completion

Hi Boris, the link leads nowhere. Where can I find the case study?