gctian's picture
update
3aa9cc0 verified
|
raw
history blame
1.91 kB
---
license: apache-2.0
task_categories:
- question-answering
language:
- zh
tags:
- comic
- llm
- text
- benchmark
- evaluate
pretty_name: comic_llm_evaluate
size_categories:
- 10K<n<100K
---
# Dataset Card for comic-eval-benchmark
<!-- Provide a quick summary of the dataset. -->
中文二次元漫画领域的基准评估数据集,包含上千部漫画作品的作者信息、画风、场景、类型、剧情等维度的选择题评估,共 41175 个单选题。
可作为二次元垂直领域大模型的评估基准。
以下是作者基于Baichuan2-13B微调的二次元领域垂直大模型,在此数据集上的评估结果:
| 模型 | zero-shot | 3-shot |
| :-----:| :----: | :----: |
| Qwen-7b | 33.647 | 36.439 |
| ChatGLM3-6b | 34.373 | 37.015 |
| BaiChuan2-13b | 37.416 | 39.08 |
| **BaiChuan2-13b-微调** | **41.035** | **41.086** |
| Yi-34b | 50.103 | 45.606 |
**欢迎贡献更多二次元领域语料及二次元大模型,如需评测请联系作者获取评测脚本。**
## Dataset Details
### Dataset Description
<!-- Provide a longer summary of what this dataset is. -->
中文二次元领域漫画基准评估数据集
### Dataset Sources
<!-- Provide the basic links for the dataset. -->
[快看漫画](https://www.kuaikanmanhua.com/)
## Dataset Structure
An example looks as follows:
<pre><code class="language-json"><span class="hljs-attr">{'question': '傲娇boss来pk漫画的作者是谁?','choices': {'A': '逐浪动漫', 'B': '王畅', 'C': '彭婷', 'D': '谷凤兰'},'answer': 'A'}</span></code></pre>
字段说明:
- question: 测试问题
- choices: ABCD四个候选选项,只有一个是正确答案
- answer: 正确选项
## Licensing Information
The dataset is available under the Apache 2.0.
## Dataset Card Authors
[快看漫画](https://www.kuaikanmanhua.com/)
## Dataset Card Contact
tgc[email protected] 整理并上传