Fetching the paper…

LV-Eval: A Balanced Long-Context Benchmark with 5 Length Levels Up to 256K · Around