[ICLR 25 Oral] RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style - View it on GitHub