So you got better results finetuning from the instruct model than the base model !?

→ View original post on X — @aravsrinivas