plaguss/vllm-structured-generation

---
size_categories: n<1K
dataset_info:
features:
- name: instruction
dtype: string
- name: structured_output
struct:
- name: format
dtype: string
- name: schema
struct:
- name: properties
struct:
- name: description
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: diet
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: habitat
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: name
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: role
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: species
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: weapon
struct:
- name: title
dtype: string
- name: type
dtype: string
- name: required
sequence: string
- name: title
dtype: string
- name: type
dtype: string
- name: generation_model
dtype: string
- name: generation
dtype: string
- name: distilabel_metadata
struct:
- name: raw_output_task
dtype: string
splits:
- name: train
num_bytes: 1688
num_examples: 2
download_size: 17403
dataset_size: 1688
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
tags:
- synthetic
- distilabel
- rlaif
---

<p align="left">
<a href="https://github.com/argilla-io/distilabel">
<img src="https://raw.githubusercontent.com/argilla-io/distilabel/main/docs/assets/distilabel-badge-light.png" alt="Built with Distilabel" width="200" height="32"/>
</a>
</p>

# Dataset Card for vllm-structured-generation

This dataset has been created with [distilabel](https://distilabel.argilla.io/).

## Dataset Summary

This dataset contains a `pipeline.yaml` which can be used to reproduce the pipeline that generated it in distilabel using the `distilabel` CLI:

```console
distilabel pipeline run --config "https://huggingface.co/datasets/plaguss/vllm-structured-generation/raw/main/pipeline.yaml"
```

or explore the configuration:

```console
distilabel pipeline info --config "https://huggingface.co/datasets/plaguss/vllm-structured-generation/raw/main/pipeline.yaml"
```

## Dataset structure

The examples have the following structure per configuration:

<details><summary> Configuration: default </summary><hr>

```json
{
"distilabel_metadata": {
"raw_output_task": "{\n \"name\": \"Thoradin Ironfist\",\n \"description\": \"Thoradin is a burly dwarf warrior, known for his unyielding determination and exceptional combat skills. He wears a set of enchanted plate armor that glows with a faint blue aura, and wields a two-handed warhammer that he can use to deliver devastating blows.\",\n \"role\": \"Tank\",\n \"weapon\": \"Two-handed warhammer + enchanted plate armor + shield + dwarf-made throwing axes for ranged attacks when necessary\"\n}"
},
"generation": "{\n \"name\": \"Thoradin Ironfist\",\n \"description\": \"Thoradin is a burly dwarf warrior, known for his unyielding determination and exceptional combat skills. He wears a set of enchanted plate armor that glows with a faint blue aura, and wields a two-handed warhammer that he can use to deliver devastating blows.\",\n \"role\": \"Tank\",\n \"weapon\": \"Two-handed warhammer + enchanted plate armor + shield + dwarf-made throwing axes for ranged attacks when necessary\"\n}",
"generation_model": "teknium/OpenHermes-2.5-Mistral-7B",
"instruction": "Generate a character from a RPG game.",
"structured_output": {
"format": "json",
"schema": {
"properties": {
"description": {
"title": "Description",
"type": "string"
},
"diet": null,
"habitat": null,
"name": {
"title": "Name",
"type": "string"
},
"role": {
"title": "Role",
"type": "string"
},
"species": null,
"weapon": {
"title": "Weapon",
"type": "string"
}
},
"required": [
"name",
"description",
"role",
"weapon"
],
"title": "Character",
"type": "object"
}
}
}
```

This subset can be loaded as:

```python
from datasets import load_dataset

ds = load_dataset("plaguss/vllm-structured-generation", "default")
```

Or simply as it follows, since there's only one configuration and is named `default`:

```python
from datasets import load_dataset

ds = load_dataset("plaguss/vllm-structured-generation")
```

</details>

size_categories: 样本规模：少于1000条
dataset_info:
数据集信息:
features:
- 字段名：instruction，数据类型：字符串
- 字段名：structured_output，为结构体类型，内部包含：
- 字段名：format，数据类型：字符串
- 字段名：schema，为结构体类型，内部包含：
- 字段名：properties，为结构体类型，其子字段包括：
- description：结构体，含title（字符串类型）、type（字符串类型）
- diet：结构体，含title（字符串类型）、type（字符串类型）
- habitat：结构体，含title（字符串类型）、type（字符串类型）
- name：结构体，含title（字符串类型）、type（字符串类型）
- role：结构体，含title（字符串类型）、type（字符串类型）
- species：结构体，含title（字符串类型）、type（字符串类型）
- weapon：结构体，含title（字符串类型）、type（字符串类型）
- 字段名：required，为字符串序列
- 字段名：title，数据类型：字符串
- 字段名：type，数据类型：字符串
- 字段名：generation_model，数据类型：字符串
- 字段名：generation，数据类型：字符串
- 字段名：distilabel_metadata，为结构体类型，内部包含raw_output_task（数据类型：字符串）
splits:
- 划分名称：train（训练集），字节大小：1688，样本数量：2
download_size: 17403
dataset_size: 1688
configs:
- 配置名称：default，数据文件：
- 对应训练集划分，路径为data/train-*
tags:
- 合成数据集（synthetic）
- distilabel
- rlaif

<p align="left">
<a href="https://github.com/argilla-io/distilabel">
<img src="https://raw.githubusercontent.com/argilla-io/distilabel/main/docs/assets/distilabel-badge-light.png" alt="基于distilabel构建" width="200" height="32"/>
</a>
</p>

# vllm-structured-generation 数据集卡片

本数据集由distilabel构建，官方地址为<https://distilabel.argilla.io/>。

## 数据集概述
本数据集包含一个`pipeline.yaml`配置文件，可通过distilabel命令行接口（CLI）运行该配置，复现生成本数据集的流水线：
console
distilabel pipeline run --config "https://huggingface.co/datasets/plaguss/vllm-structured-generation/raw/main/pipeline.yaml"

或查看该配置的详情：
console
distilabel pipeline info --config "https://huggingface.co/datasets/plaguss/vllm-structured-generation/raw/main/pipeline.yaml"

## 数据集结构
每个配置下的数据样本具有如下结构：

<details><summary> 配置：default </summary><hr>

json
{
"distilabel_metadata": {
"raw_output_task": "{
"name": "Thoradin Ironfist",
"description": "Thoradin is a burly dwarf warrior, known for his unyielding determination and exceptional combat skills. He wears a set of enchanted plate armor that glows with a faint blue aura, and wields a two-handed warhammer that he can use to deliver devastating blows.",
"role": "Tank",
"weapon": "Two-handed warhammer + enchanted plate armor + shield + dwarf-made throwing axes for ranged attacks when necessary"
}"
},
"generation": "{
"name": "Thoradin Ironfist",
"description": "Thoradin is a burly dwarf warrior, known for his unyielding determination and exceptional combat skills. He wears a set of enchanted plate armor that glows with a faint blue aura, and wields a two-handed warhammer that he can use to deliver devastating blows.",
"role": "Tank",
"weapon": "Two-handed warhammer + enchanted plate armor + shield + dwarf-made throwing axes for ranged attacks when necessary"
}",
"generation_model": "teknium/OpenHermes-2.5-Mistral-7B",
"instruction": "Generate a character from a RPG game.",
"structured_output": {
"format": "json",
"schema": {
"properties": {
"description": {
"title": "Description",
"type": "string"
},
"diet": null,
"habitat": null,
"name": {
"title": "Name",
"type": "string"
},
"role": {
"title": "Role",
"type": "string"
},
"species": null,
"weapon": {
"title": "Weapon",
"type": "string"
}
},
"required": [
"name",
"description",
"role",
"weapon"
],
"title": "Character",
"type": "object"
}
}
}

该子集可通过如下方式加载：
python
from datasets import load_dataset

ds = load_dataset("plaguss/vllm-structured-generation", "default")

由于本数据集仅包含一个名为default的配置，也可通过如下简化方式加载：
python
from datasets import load_dataset

ds = load_dataset("plaguss/vllm-structured-generation")

</details>

原始信息汇总

数据集概述

数据集结构

特征

• instruction: 类型为字符串。
• structured_output: 结构化输出，包含以下子特征：
  • format: 类型为字符串。
  • schema: 包含以下子特征：
    • properties: 包含以下子特征：
      • description: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • diet: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • habitat: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • name: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • role: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • species: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
      • weapon: 包含以下子特征：
        • title: 类型为字符串。
        • type: 类型为字符串。
    • required: 类型为字符串序列。
    • title: 类型为字符串。
    • type: 类型为字符串。
• generation_model: 类型为字符串。
• generation: 类型为字符串。
• distilabel_metadata: 包含以下子特征：
  • raw_output_task: 类型为字符串。

数据集划分

• train: 包含2个样本，占用1688字节。

数据集大小

• 下载大小: 17403字节
• 数据集大小: 1688字节

配置

• default: 包含一个训练集，路径为data/train-*。

标签

• synthetic
• distilabel
• rlaif