DavidLanz/alpaca-tw-input-output-52k

---
task_categories:
- text-generation
- conversational
- question-answering
language:
- en
size_categories:
- 10K<n<100K
license: cc-by-4.0
tags:
- gpt3
- alpaca
- instruction-finetuning
---
# Dataset Card for "alpaca-tw-input-output-52k"

This dataset contains English Instruction-Following generated by GPT-3.5 using Alpaca prompts for fine-tuning LLMs.

The dataset was originaly shared in this repository: https://github.com/ntunlplab/traditional-chinese-alpaca. This is just a wraper for compatibility with huggingface's datasets library.

## Dataset structure

It contains 52K instruction-following data generated by GPT-3.5 using the same prompts as in Alpaca.
The dataset has the same format as Alpaca data, except the output is generated by GPT-3.5:

- `instruction`: `str`, describes the task the model should perform. Each of the 52K instructions is unique.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, the answer to the instruction as generated by `GPT-3.5`.

## Difference with the original Alpaca dataset

The original Alpaca dataset used text-davinci-003 to complete the prompts. This dataset uses those same prompts, but generating the completions with GPT-3.5. Thus, in general, the responses are of higher quality and lenght. Here is an example:

#### Example from Alpaca-GPT3:

```bash
{
'instruction': '辨識那一個平台與其他不同。',
'input': 'Twitter, Instagram, Telegram',
'output': '在Twitter、Instagram和Telegram之間，Telegram是與其他兩者最不同的平台。'
}
```
## Licensing Information

The dataset is available under the [Creative Commons NonCommercial (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/legalcode).

原始信息汇总

数据集卡片 "alpaca-tw-input-output-52k"

概述

该数据集包含由GPT-3.5生成的英语指令遵循数据，使用Alpaca提示进行LLM微调。

数据集结构

数据集包含52K由GPT-3.5生成的指令遵循数据，使用与Alpaca相同的提示。数据格式与Alpaca数据相同，除了输出是由GPT-3.5生成的：

• instruction: str, 描述模型应执行的任务。52K条指令每条都是唯一的。
• input: str, 任务的可选上下文或输入。
• output: str, GPT-3.5生成的指令答案。

与原始Alpaca数据集的区别

原始Alpaca数据集使用text-davinci-003完成提示。该数据集使用相同的提示，但使用GPT-3.5生成完成。因此，总体上，响应的质量和长度更高。

许可信息

该数据集在Creative Commons NonCommercial (CC BY-NC 4.0)许可下可用。