[EVAL] Add IFBench

[lewtun]

Evaluation short description

IFEval is widely used to measure the instruction following capabilities of LLMs but is now quite saturated. IFBench is an improved version of IFEval which has broader diversity of task constraints. I think IFBench is likely to provide better signal for the next generation of models.

Evaluation metadata

Provide all available

• Paper url: https://arxiv.org/abs/2507.02833
• Github url: https://github.com/allenai/IFBench
• Dataset url: https://huggingface.co/collections/allenai/ifbench-683f590687f61b512558cdf1