melvindave/legal-synth-docs

Legal Synth Docs — LLM-as-Judge数据集是用于微调大型语言模型（LLM）以评估合同条款质量的训练数据，评估维度包括清晰度、完整性、风险平衡和可执行性（每项0-25分，总分0-100分）。数据集包含277个标注的条款示例，覆盖60多种条款类型，如付款、终止、赔偿、知识产权转让、适用法律、保密协议、服务级别协议、不可抗力、竞业禁止、数据保护、保修、责任、特许经营、雇佣、软件许可、并购、房地产等。数据集还提供了评分标准和字段描述，包括指令、输入和输出字段。

Legal Synth Docs — LLM-as-Judge Dataset is training data for fine-tuning an LLM to evaluate contract clause quality across four dimensions: clarity, completeness, risk_balance, and enforceability (0–25 each, total 0–100). The dataset contains 277 annotated clause examples spanning the full quality spectrum (scores 15–95/100), covering 60+ clause types including payment, termination, indemnification, IP assignment, governing law, NDA, SLA, force majeure, non-compete, data protection, warranty, liability, franchise, employment, software licensing, M&A, real estate, and more. The dataset also provides a scoring rubric and field descriptions, including instruction, input, and output fields.