kurakurai/IFEval-FR
收藏资源简介:
IFEval-FR是M-IFEval指令跟随基准的法语版本,格式与google/IFEval数据集相同。该版本经过Claude Opus 4.8的审查和清理,修复了提示/验证器不匹配、偏移约束、未翻译提示以及语法/格式问题。数据集包含235个提示,仅有一个训练分割,字段包括唯一ID(key)、法语指令(prompt)、可验证指令列表(instruction_id_list,以fr:为前缀)以及每个指令的参数(kwargs)。它用于评估模型在法语环境下的指令跟随能力,适用于文本生成任务,采用Apache-2.0许可证。
IFEval-FR is the French version of the M-IFEval instruction-following benchmark, formatted identically to the google/IFEval dataset. This version has been reviewed and cleaned by Claude Opus 4.8, fixing issues including prompt/validator mismatches, offset constraints, untranslated prompts, and grammatical/formatting problems. The dataset contains 235 prompts, with only a single training split. Its fields include unique ID (key), French instruction (prompt), verifiable instruction list (instruction_id_list, prefixed with fr:), and parameters (kwargs) for each instruction. It is used to evaluate models' instruction-following capabilities in a French-language context, is suitable for text generation tasks, and is released under the Apache-2.0 license.




