anon-neurips-ed-2026/polymorphic-sybil-benchmark-data

Polymorphic Sybil Benchmark: Retrieval Indexes数据集是Polymorphic Sybil检索中毒基准的配套检索工具，包含大型预构建索引，适用于检索增强生成任务。数据集基于多个开放域QA数据集构建，使用Wikipedia DPR 100-word split作为语料库。内容包括BM25和E5 FAISS索引，分别用于不同的检索和重排序流程。数据集遵循CC BY-SA 4.0许可，继承自Wikipedia、NQ、HotpotQA等上游许可。

The Polymorphic Sybil Benchmark: Retrieval Indexes dataset serves as a companion to the Polymorphic Sybil Retrieval Poisoning Benchmark, providing large pre-built indexes for retrieval-augmented generation tasks. It is built upon multiple open-domain QA datasets and uses the Wikipedia DPR 100-word split as its corpus. The dataset includes BM25 and E5 FAISS indexes for different retrieval and reranking pipelines. It is licensed under CC BY-SA 4.0, inheriting from upstream licenses such as Wikipedia, NQ, and HotpotQA.