jfbench-verified
收藏Hugging Face2026-03-16 更新2026-03-20 收录
下载链接:
https://huggingface.co/datasets/pfnet/jfbench-verified
下载链接
链接失效反馈官方服务:
资源简介:
该数据集包含用于评估生成式AI模型日语指令跟随性能的已验证数据。数据集特征包括约束条件、数据ID、提示、提示文档、提示来源和响应。其中,提示基于jfbench基准套件构建,提示文档最初来自IFBench_test数据集,响应由gpt-oss-120b模型生成,并通过jfbench基准套件验证。数据集包含800个测试样本,采用ODC-BY-1.0许可协议,适用于研究和教育用途。
This dataset contains verified data for evaluating the Japanese instruction-following performance of generative AI models. Its features include constraints, data ID, prompts, prompt documents, prompt sources, and responses. The prompts are constructed based on the jfbench benchmark suite, with the prompt documents originally sourced from the IFBench_test dataset. The responses were generated by the gpt-oss-120b model and verified using the jfbench benchmark suite. The dataset consists of 800 test samples and is licensed under ODC-BY-1.0, intended for research and educational purposes.
提供机构:
Preferred Networks, Inc.
创建时间:
2026-03-16



