[2/2] 2) Instruction finetuning: usually RLHF to generate a specific response. (But can also refer to 1b depending on the technique) 3) Long-context finetuning: like pretraining but w longer contexts (as per Code Llama paper)?

→ View original post on X — @rasbt