D2P
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/QiYao-Wang/AutoPatent
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为D2P基准，包含了1500组草稿与专利的配对样本，这些样本被用于训练AutoPatent框架，旨在根据初步草稿生成完整的专利文档。此外，该数据集还用于对包括LLAMA3.1-8B、Qwen2.5-7B和Mistral-7B在内的多种模型进行完全监督的微调。规模上，该数据集包含了1500组草稿与专利的配对，其任务是专利生成。
This dataset, termed the D2P Benchmark, comprises 1500 paired samples of drafts and patent documents. It is utilized to train the AutoPatent framework, which is designed to generate complete patent documents from preliminary drafts. Additionally, this dataset supports fully-supervised fine-tuning of multiple large language models including LLAMA3.1-8B, Qwen2.5-7B and Mistral-7B. The core task of this benchmark is patent generation.
提供机构：
AutoPatent framework authors
