Smail/bespoke-manim-preprocessed

这是一个经过筛选和格式化的bespoke-manim数据集的版本，专门设计用于微调遵循指令的语言模型，以从丰富的多模态提示生成可执行Manim代码。原始数据集是为了促进使用Manim生成数学动画而策划的。这个版本仅保留有效的条目（无执行错误，附带视频），并将数据重构为适合监督微调的指令-输出对。

This dataset is a filtered and reformatted version of the bespoke-manim dataset, designed specifically for fine-tuning instruction-following language models to generate executable Manim code from rich multimodal prompts. The original dataset was curated to facilitate the generation of mathematical animations using Manim. This version retains only valid entries (no execution errors, with attached video) and restructures the data into instruction-output pairs suitable for supervised fine-tuning.