oliverkinch/da-instruct-dynaword-contemporary

da-instruct-dynaword-contemporary 是一个丹麦语指令微调数据集，通过反向翻译从 danish-dynaword 数据集生成，仅限于当代丹麦语子集，未进行基于注释的质量过滤。每行数据是一个 (prompt, target) 对，其中 target 是来自精选 DynaWord 子集的真实丹麦语文本段落，prompt 是一个现实的丹麦用户指令，可以合理地引出语言模型生成该文本。prompt 由 Qwen/Qwen3.5-397B-A17B 模型使用反向翻译方法生成：给定一个段落及其源上下文，模型被要求编写一个用户消息，该消息会导致聊天机器人产生类似的文本。特定领域的提示模板（政府、文学、学术、税务指导、新闻、演讲）确保生成的指令与源文本的语域和体裁匹配。数据集构建包括子集选择、段落提取、提示生成和结果统计，共从 32 个 dynaword 子集的 10,000 个随机采样源 ID 中生成了 8,271 个指令对。

da-instruct-dynaword-contemporary is a Danish instruction fine-tuning dataset generated via backtranslation from the danish-dynaword dataset, restricted to contemporary Danish subsets with no annotation-based quality filtering. Each row is a (prompt, target) pair where target is a passage of authentic Danish text drawn from a curated subset of DynaWord, and prompt is a realistic Danish user instruction that would plausibly elicit that text from a language model. Prompts were generated by Qwen/Qwen3.5-397B-A17B using a backtranslation approach: given a passage and its source context, the model is asked to write the user message that would cause a chatbot to produce a similar text. Domain-specific prompt templates (government, literary, academic, tax guidance, news, speech) ensure that the generated instructions match the register and genre of the source text. Dataset construction includes subset selection, passage extraction, prompt generation, and results, with 8,271 instruction pairs generated from 10,000 randomly sampled source IDs across 32 dynaword subsets.