14890fad56
* feat: add agents/actions/write_markdown * [ADD] add evaluation result of base model on 5/10 epochs * Rename mother.json to mother_v1_2439.json * Add files via upload * [DOC] update README * Update requirements.txt update mpi4py installation * Update README_EN.md update English comma * Update README.md 基于母亲角色的多轮对话模型微调完毕。已上传到 Huggingface。 * 多轮对话母亲角色的微调的脚本 * Update README.md 加上了王几行XING 和 思在 的作者信息 * Update README_EN.md * Update README.md * Update README_EN.md * Update README_EN.md * Changes to be committed: modified: .gitignore modified: README.md modified: README_EN.md new file: assets/EmoLLM_transparent.png deleted: assets/Shusheng.jpg new file: assets/Shusheng.png new file: assets/aiwei_demo1.gif new file: assets/aiwei_demo2.gif new file: assets/aiwei_demo3.gif new file: assets/aiwei_demo4.gif * Update README.md rectify aiwei_demo.gif * Update README.md rectify aiwei_demo style * Changes to be committed: modified: README.md modified: README_EN.md * Changes to be committed: modified: README.md modified: README_EN.md * [Doc] update readme * [Doc] update readme * Update README.md * Update README_EN.md * Update README.md * Update README_EN.md * Delete datasets/mother_v1_2439.json * Rename mother_v2_3838.json to mother_v2.json * Delete datasets/mother_v2.json * Add files via upload * Update README.md * Update README_EN.md * [Doc] Update README_EN.md minor fix * InternLM2-Base-7B QLoRA微调模型 链接和测评结果更新 * add download_model.py script, automatic download of model libraries * 清除图片的黑边、更新作者信息 modified: README.md new file: assets/aiwei_demo.gif deleted: assets/aiwei_demo1.gif modified: assets/aiwei_demo2.gif modified: assets/aiwei_demo3.gif modified: assets/aiwei_demo4.gif * rectify aiwei_demo transparent * transparent * modify: aiwei_demo table--->div * modified: aiwei_demo * modify: div ---> table * modified: README.md * modified: README_EN.md * update model config file links * Create internlm2_20b_chat_lora_alpaca_e3.py 20b模型的配置文件 * update model config file links update model config file links * Revert "update model config file links" --------- Co-authored-by: jujimeizuo <fengzetao.zed@foxmail.com> Co-authored-by: xzw <62385492+aJupyter@users.noreply.github.com> Co-authored-by: Zeyu Ba <72795264+ZeyuBa@users.noreply.github.com> Co-authored-by: Bryce Wang <90940753+brycewang2018@users.noreply.github.com> Co-authored-by: zealot52099 <songyan5209@163.com> Co-authored-by: HongCheng <kwchenghong@gmail.com> Co-authored-by: Yicong <yicooong@qq.com> Co-authored-by: Yicooong <54353406+Yicooong@users.noreply.github.com> Co-authored-by: aJupyter <ajupyter@163.com> Co-authored-by: MING_X <119648793+MING-ZCH@users.noreply.github.com> Co-authored-by: Ikko Eltociear Ashimine <eltociear@gmail.com> Co-authored-by: HatBoy <null2none@163.com> Co-authored-by: ZhouXinAo <142309012+zxazys@users.noreply.github.com>
72 lines
3.6 KiB
Markdown
72 lines
3.6 KiB
Markdown
# EmoLLM数据集
|
||
|
||
* 数据集按用处分为两种类型：**General** 和 **Role-play**
|
||
* 数据按格式分为两种类型：**QA** 和 **Conversation**
|
||
* 数据汇总：General（**6个数据集**）；Role-play（**5个数据集**）
|
||
|
||
## 数据集类型
|
||
|
||
* **General**：通用数据集，包含心理学知识、心理咨询技术等通用内容
|
||
* **Role-play**：角色扮演数据集，包含特定角色对话风格数据等内容
|
||
|
||
## 数据类型
|
||
|
||
* **QA**：问答对
|
||
* **Conversation**：多轮对话
|
||
|
||
## 数据集汇总
|
||
|
||
| Category | Dataset | Type | Total |
|
||
| :---------: | :-------------------: | :----------: | :-----: |
|
||
| *General* | data | Conversation | 5600+ |
|
||
| *General* | data_pro | Conversation | 36,500+ |
|
||
| *General* | multi_turn_dataset_1 | Conversation | 36,000+ |
|
||
| *General* | multi_turn_dataset_2 | Conversation | 27,000+ |
|
||
| *General* | single_turn_dataset_1 | QA | 14,000+ |
|
||
| *General* | single_turn_dataset_2 | QA | 18,300+ |
|
||
| *Role-play* | aiwei | Conversation | 4000+ |
|
||
| *Role-play* | SoulStar | QA | 11,200+ |
|
||
| *Role-play* | tiangou | Conversation | 3900+ |
|
||
| *Role-play* | mother | Conversation | 40,300+ |
|
||
| *Role-play* | scientist | Conversation | 28,400+ |
|
||
| …… | …… | …… | …… |
|
||
|
||
## 数据集来源
|
||
|
||
### **General**
|
||
|
||
* 数据集 `data` 来自本项目
|
||
* 数据集 `data_pro` 来自本项目
|
||
* 数据集 `multi_turn_dataset_1` 来源 [Smile](https://github.com/qiuhuachuan/smile)
|
||
* 数据集 `multi_turn_dataset_2` 来源 [CPsyCounD](https://github.com/CAS-SIAT-XinHai/CPsyCoun)
|
||
* 数据集 `single_turn_dataset_1` 来自本项目
|
||
* 数据集 `single_turn_dataset_2` 来自本项目
|
||
|
||
### **Role-play**
|
||
|
||
* 数据集 `aiwei` 来自本项目
|
||
* 数据集 `tiangou` 来自本项目
|
||
* 数据集 `SoulStar` 来源 [SoulStar](https://github.com/Nobody-ML/SoulStar)
|
||
* 数据集 `mother` 来自本项目
|
||
* 数据集 `scientist` 来自本项目
|
||
|
||
## 数据集去重
|
||
|
||
结合绝对匹配以及模糊匹配(Simhash)算法，对数据集进行去重以提升微调模型的效果。在确保数据集的高质量的同时，通过调整阈值减少因错误匹配而丢失重要数据的风险。
|
||
|
||
### **Simhash算法介绍**
|
||
|
||
Simhash（相似性哈希）是一种用于检测大量数据中相似或重复项的算法。它通过将文本转换为一组数值指纹来工作，这些指纹对相似的文本具有高度的相似性。Simhash算法对于处理文本数据特别有效，尤其是在处理大量数据时。
|
||
|
||
### **Simhash实现步骤**
|
||
|
||
*文本预处理：将文本数据转换为适合Simhash处理的格式。这可能包括分词、去除停用词、词干提取等。
|
||
*生成Simhash指纹：对预处理后的文本应用Simhash算法，生成一组数值指纹。每个指纹代表文本内容的一个哈希值。
|
||
*比较指纹：通过比较哈希值的相似性来识别重复或相似的记录。Simhash的特点是即使在文本有少量差异时，生成的哈希值也具有较高的相似性。
|
||
*确定阈值：设置一个相似性阈值，只有当两个指纹的相似度超过这个阈值时，才认为它们代表相似或重复的记录。
|
||
*处理相似记录：对于被标记为相似的记录，可以进一步人工审查或自动合并，以消除重复。
|
||
|
||
### deduplicate.py用法
|
||
|
||
`deduplicate.py` 用于将datasets下以模型命名的文件夹下(例如：'datasets/qwen').json数据进行去重，输出去重后的数据到 `datasets/qwen/dedup` 文件夹下。
|