easyr1-10k-hard-qwen7b-easy-gta1-4MP-synthetic-prompts-qwen25vl7binstruct

# easyr1-10k-hard-qwen7b-easy-gta1-4MP-synthetic-prompts-qwen25vl7binstruct

This dataset is derived from [mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP](https://huggingface.co/datasets/mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP)
with synthetic prompts generated using Qwen/Qwen2.5-VL-7B-Instruct.

## Generation Details

- **Generated on**: 2025-08-21 14:56:03 UTC
- **Source dataset**: mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP
- **Split processed**: train
- **Model used**: Qwen/Qwen2.5-VL-7B-Instruct
- **Total entries**: 10000

## Synthetic Prompt Generation

Each entry in the original dataset has been processed to generate synthetic prompts that represent
possible user tasks. The synthetic prompts are generated based on:

1. The UI element that was clicked
2. The context visible in the screenshot
3. Common user intentions for similar UI interactions

## New Fields Added

- `synthetic_prompts`: List of generated synthetic prompts (typically 3)
- `original_prompt`: The original user instruction before synthesis
- `selected_synthetic_prompt`: The synthetic prompt selected for the updated messages

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP-synthetic-prompts-qwen25vl7binstruct")

# Access the data
sample = dataset['train'][0]

# Original prompt
print(sample['original_prompt'])

# All generated synthetic prompts
print(sample['synthetic_prompts'])

# The selected synthetic prompt used in messages
print(sample['selected_synthetic_prompt'])

# Updated messages with synthetic prompt
print(sample['messages'])
```

## License

Inherits the license from the original dataset: mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP

# easyr1-10k-hard-qwen7b-easy-gta1-4MP-synthetic-prompts-qwen25vl7binstruct

本数据集衍生自 [mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP](https://huggingface.co/datasets/mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP)，其合成提示词由 Qwen/Qwen2.5-VL-7B-Instruct 生成。

## 生成详情
- **生成时间**：2025-08-21 14:56:03 UTC
- **源数据集**：mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP
- **处理拆分集**：训练集（train）
- **使用模型**：Qwen/Qwen2.5-VL-7B-Instruct
- **总条目数**：10000

## 合成提示词生成
本数据集的原始条目均经过处理，以生成可表征实际用户任务的合成提示词。合成提示词的生成依据如下：
1. 被点击的UI元素
2. 截图中可见的上下文信息
3. 同类UI交互场景下的常见用户意图

## 新增字段
- `synthetic_prompts`：生成的合成提示词列表（通常包含3条）
- `original_prompt`：合成前的原始用户指令
- `selected_synthetic_prompt`：用于更新对话消息的选定合成提示词

## 使用方法
python
from datasets import load_dataset

dataset = load_dataset("mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP-synthetic-prompts-qwen25vl7binstruct")

# 访问数据样本
sample = dataset['train'][0]

# 原始提示词
print(sample['original_prompt'])

# 所有生成的合成提示词
print(sample['synthetic_prompts'])

# 用于对话消息的选定合成提示词
print(sample['selected_synthetic_prompt'])

# 带有合成提示词的更新后对话消息
print(sample['messages'])

## 许可证
本数据集继承源数据集 mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP 的许可证协议。

搜集汇总

背景与挑战

背景概述

该数据集源自easyr1-10k-hard-qwen7b-easy-gta1-4MP，通过Qwen2.5-VL-7B-Instruct模型生成了合成提示，用于模拟用户任务。它包含10000个条目，新增了合成提示列表、原始提示和选定提示等字段，适用于数据加载和处理。