Interesting point about backend variance. Do you think serving layer should become part of standard LLM eval reporting?
Interesting point about backend variance. Do you think serving layer should become part of standard LLM eval reporting?