NullRabbit/nr-bundles-public

nr-bundles-public 是一个精心策划的多模态区块链验证器基础设施数据集，涵盖攻击和良性工作负载。该数据集包含31个bundle，涉及两个区块链（Sui和Solana）、七个漏洞家族和十个攻击原语。作为NullRabbit的bundle语料库的第一个公开样本，它支持对去中心化网络自主防御的机器学习研究。每个bundle是一个多模态观察，包括网络包、主机遥测、应用指标和观察响应，采用链无关的模式。数据集旨在用于多模态异常检测研究、对抗鲁棒性基准测试、格式采用和方法论研究。数据集在受控的实验室环境中捕获，仅包含Sui和Solana链，是完整语料库的精选子集，具有局限性，如原始网络包文件被删除。使用数据集需要通过snapshot_download和bundle-spec解析器，而不是标准的HF加载器。nr-bundles-public is a curated, multi-modal dataset of blockchain validator infrastructure under attack and benign workloads. It includes thirty-one bundles across two chains (Sui, Solana), seven vulnerability families, and ten attack primitives. Released as the first public sample of NullRabbits bundle corpus, it underpins ML research on autonomous defence for decentralised networks. Each bundle is a multi-modal observation capturing network packets, host telemetry, application metrics, and observed responses under a chain-agnostic schema. The dataset is intended for research on multi-modal anomaly detection, benchmarking adversarial robustness, format adoption, and methodology study. It is captured in a controlled lab environment, limited to two chains, and is a curated subset with limitations such as dropped raw packet captures. Usage requires snapshot_download and bundle-spec parsers rather than standard HF loaders.