text-vision-audio-2k-test
收藏资源简介:
这是一个包含2000个样本的多模态(文本+图片+音频)测试数据集,用于测试兼容HF的处理器apply_chat_template功能。每个样本包含用户和助手两个角色的对话,对话内容包括图片、音频和文本。数据集分为训练集,可以通过特定的方式加载。
This is a multimodal (text + image + audio) test dataset containing 2000 samples, designed to test the `apply_chat_template` function of HF-compatible processors. Each sample consists of a dialogue between the user and assistant roles, with the dialogue content including images, audio and text. This dataset is split into training subsets and can be loaded via a specific approach.
数据集概述
基本信息
- 数据集名称: text-vision-audio-2k-test
- 用途: 用于测试多模态(文本+视觉+音频)格式的2k样本数据集
- 兼容性: 兼容Hugging Face的
apply_chat_template处理器
数据集结构
- 特征:
messages:content:path(string): 文件路径text(string): 文本内容type(string): 内容类型(如image、audio、text)
role(string): 角色(如user、assistant)
- 数据分割:
train:- 字节数: 526000
- 样本数: 2000
- 下载大小: 5957
- 数据集大小: 526000
下载与加载
-
加载方式 (Axolotl): yaml datasets:
- path: Nanobit/text-vision-audio-2k-test type: chat_template
-
额外文件下载: bash wget https://huggingface.co/datasets/Nanobit/text-vision-audio-2k-test/resolve/main/African_elephant.jpg wget https://huggingface.co/datasets/Nanobit/text-vision-audio-2k-test/resolve/main/En-us-African_elephant.oga
样本格式
每个样本重复2000次,格式如下: py [ { "role": "user", "content": [ {"type": "image", "path": LOCAL_IMAGE_PATH}, {"type": "audio", "path": LOCAL_AUDIO_PATH}, { "type": "text", "text": "How would you summarize the above image and audio in one phrase?", }, ], }, { "role": "assistant", "content": [ { "type": "text", "text": "The image and audio is about an African elephant.", } ], }, ]
数据来源
- 图片来源: https://upload.wikimedia.org/wikipedia/commons/e/ec/African_elephant.jpg
- 音频来源: https://upload.wikimedia.org/wikipedia/commons/a/ad/En-us-African_elephant.oga




