TylerHMC/MIDICaps-500-label

该数据集包含MIDI音乐文件的元数据，每个样本包含token化的输入序列、token数量、块索引、块总数、嵌入向量（5个1024维浮点向量）及其掩码、流派、情绪、乐器家族、MIDI文件路径和源文件路径。数据集共840816个训练样本，适用于音乐信息检索、音乐生成、风格分类等任务。This dataset contains metadata of MIDI music files. Each sample includes tokenized input IDs, token count, chunk index, number of chunks, embeddings (5 vectors of 1024 floats) with masks, genre, mood, instrument families, MIDI path, and source file path. The dataset has 840,816 training examples, suitable for tasks such as music information retrieval, music generation, and style classification.