achinta3/cybersec-jsonschemabench

这是一个小型的JSONSchemaBench风格网络安全基准测试，基于当前仓库中的规范化演示数据构建。每行数据包括unique_id、json_schema、prompt、input_jsonl和ground_truth_json。当前版本为0.2.0，总共有100行数据，来源包括flaws_cloudtrail和suricata_eve。主要用于评估模型在给定prompt和input_jsonl下的输出，解析为JSON后验证其是否符合json_schema，并与ground_truth_json进行评分。This is a small JSONSchemaBench-style cybersecurity benchmark built from the current normalized demo data in this repo. Each row includes unique_id, json_schema, prompt, input_jsonl, and ground_truth_json. The current version is 0.2.0 with a total of 100 rows, sourced from flaws_cloudtrail and suricata_eve. It is intended for evaluating model outputs given prompt and input_jsonl, parsing them as JSON, validating against json_schema, and scoring against ground_truth_json.