Update README.md
Browse files
README.md
CHANGED
@@ -5,12 +5,19 @@ size_categories:
|
|
5 |
language:
|
6 |
- en
|
7 |
---
|
|
|
|
|
8 |
* English multi document Q&A data created using RedPajamaCommonCrawl data as reference text
|
9 |
-
*
|
10 |
color=red> one reference document, 199 irrelevant documents, and a Q-A pair based on the reference document</font>. It can be used to train models to extract the target information from a large number of documents.
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
* 以RedPajamaCommonCrawl数据为参考文本,制作的英文多文档问答数据
|
13 |
-
*
|
14 |
-
|
15 |
|
16 |
-
* dataset size:
|
|
|
5 |
language:
|
6 |
- en
|
7 |
---
|
8 |
+
* Update on December 24, 2023: Improve the format of answers and require all answers to be provided in their original text.
|
9 |
+
|
10 |
* English multi document Q&A data created using RedPajamaCommonCrawl data as reference text
|
11 |
+
* In the Raw dataset, each sample contains <font
|
12 |
color=red> one reference document, 199 irrelevant documents, and a Q-A pair based on the reference document</font>. It can be used to train models to extract the target information from a large number of documents.
|
13 |
+
* After filtering, integrating, and transforming the raw data into chatml format instruction fine-tuning data, each sample contains approximately 30 reference documents and 5 corresponding QA pairs.
|
14 |
+
|
15 |
+
<br/>
|
16 |
+
|
17 |
+
* 2023.12.4更新:改进答案的格式,强制所有答案在回答时必须给出原文。
|
18 |
|
19 |
* 以RedPajamaCommonCrawl数据为参考文本,制作的英文多文档问答数据
|
20 |
+
* 原始数据中,每个样本包含 <font color=red> 一个参考文档、199个无关文档、一个基于参考文档的问答对</font>。可以训练模型从大量文档中抽取关键信息的能力。
|
21 |
+
* 原始数据经过筛选、整合转化为chatml形式的指令微调数据后,每条数据大约包含30个参考文档,以及5个对应的问答对。
|
22 |
|
23 |
+
* dataset size: 11k
|