Best of N benchmark

[natolambert]

1. Take a few chat models as the “base set”, say 1-3, like tulu 2 7b and tulu 2 13b (maybe olmo-instruct)
2. Generate ~8 completions per prompt in AlpacaEval (this is the heldout set)
3. Use each RM to choose the best-of-1 from that set, then run alpaca eval on the outputs
4. Score the delta for each RM in the batch on a set task (alpacaeval) and set base model (tulu)
5. Could do this with MTBench, but two turn is harder

Obvi flaws, but that seems WAY better than nothing.

[natolambert]

@yuchenlin is starting this, woohoo!

[natolambert]

Partially closed in #30 , wrapping up soon.