Hi lm-eval team,
I am wondering if there are plans to add LongBench v2, Babilong, InfiniteBench, and Phonebook datasets to the evaluation tasks? These are useful for long-context LLM evaluation.

• LongBench v2: 2nd-gen long-text benchmark with 20 tasks (8k–2M words), focusing on real-world deep reasoning. https://github.com/THUDM/LongBench
• Babilong: Focuses on ultra-long context reasoning (up to 10M tokens) with 20 tasks, testing fact inference across extended documents.https://github.com/booydar/babilong
• InfiniteBench: A long-context suite (100k+ tokens) with 12 tasks (retrieval, math, code, etc.) for real/scenario-based evaluation.https://github.com/OpenBMB/InfiniteBench

Thanks!