oumi-ai/oumi-c2d-d2c-subset
收藏Hugging Face2025-04-04 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/oumi-ai/oumi-c2d-d2c-subset
下载链接
链接失效反馈官方服务：
资源简介：
oumi-c2d-d2c-subset是一个为微调语言模型进行声明验证而设计的文本数据集。该数据集从C2D-and-D2C-MiniCheck训练集中提取提示，并由Llama-3.1-405B-Instruct模型生成响应。oumi-c2d-d2c-subset被用于训练HallOumi-8B模型，该模型以77.2%的宏F1得分超过了Claude Sonnet 3.5、OpenAI o1等最新技术水平。
oumi-c2d-d2c-subset is a text dataset designed for fine-tuning language models for claim verification. The prompts were sourced from the C2D-and-D2C-MiniCheck training sets and the responses were generated by the Llama-3.1-405B-Instruct model. oumi-c2d-d2c-subset was used to train the HallOumi-8B model, which achieved a 77.2% Macro F1 score, surpassing SOTA models like Claude Sonnet 3.5 and OpenAI o1.
提供机构：
oumi-ai
