LongBench-v2
A benchmark dataset for evaluating large language models on long-context understanding. It comprises multiple-choice long-context question answering tasks and serves as a standard evaluation set in long-context QA research.
Timeline (last 90 days)
Under an 18k character budget, anchored hybrid retrieval achieved 36.18% accuracy, BM25 achieved 35.98%, while the best direct planner-guided method reached only 34.19%.
The research used all 503 multiple-choice questions from the LongBench-v2 dataset as the primary diagnostic set, with Qwen2.5-7B-Instruct as the answer model.
The study's limitations include being restricted to a single model, Qwen2.5-7B-Instruct, and a single dataset, LongBench-v2, with a small amount of training data (only 168 questions).
All Facts (3)
Under an 18k character budget, anchored hybrid retrieval achieved 36.18% accuracy, BM25 achieved 35.98%, while the best direct planner-guided method reached only 34.19%.
50%UnverifiedThe research used all 503 multiple-choice questions from the LongBench-v2 dataset as the primary diagnostic set, with Qwen2.5-7B-Instruct as the answer model.
50%UnverifiedThe study's limitations include being restricted to a single model, Qwen2.5-7B-Instruct, and a single dataset, LongBench-v2, with a small amount of training data (only 168 questions).
50%