somukandula/prompt-router-dataset

该数据集包含700个合成的用户提示，用于训练和评估提示路由分类器。这些提示被标记为路由到四个模型层级之一：cheap_small_text（用于简单的文本问答、重写、摘要和简单聊天）、code_model（用于编码、调试、代码解释和开发运维）、vision_model（用于涉及图像、截图、图表和图示的提示）和strong_general（用于复杂推理、长上下文、研究和规划）。数据集分为训练集（560个提示）和测试集（140个提示），每个示例包含唯一的标识符、提示文本、标签和分割信息。

This dataset contains prompts labeled for routing to one of four model tiers: cheap_small_text (for simple text QA, rewriting, summarization, easy chat), code_model (for coding, debugging, code explanation, devops), vision_model (for prompts involving images, screenshots, charts, diagrams), and strong_general (for complex reasoning, long context, research, planning). The dataset is split into a training set (560 prompts) and a test set (140 prompts), with each example containing a unique identifier, the prompt text, a label, and the split information.