SCBench
收藏资源简介:
SCBench(SharedContextBench)是一个综合基准,用于评估在KV缓存为中心的角度下,高效长上下文方法的性能。该基准涵盖了12个多样化的任务,测试了四个关键的长上下文能力:字符串检索、语义检索、全局信息处理和多任务处理。数据集包含多个配置,如multi_turn_choice_eng、multi_turn_kv等,每个配置都有特定的特征,如上下文、多轮对话、ID等。数据集提供了训练集的大小和样本数量。
SCBench (SharedContextBench) is a comprehensive benchmark for evaluating the performance of efficient long-context methods from a KV cache-centric perspective. This benchmark covers 12 diverse tasks and tests four core long-context capabilities: string retrieval, semantic retrieval, global information processing, and multi-task processing. The dataset includes multiple configurations such as multi_turn_choice_eng, multi_turn_kv, etc., each with specific characteristics including context, multi-turn dialogues, IDs, and more. The dataset provides the size of the training set and the number of samples.




