Molmo2-MultiImageQA

# Molmo2-MultiImageQA
Molmo2-MultiImageQA is a **multi-image extension** of [PixMo-AskModelAnything](https://huggingface.co/datasets/allenai/pixmo-ask-model-anything), an instruction-tuning dataset for vision-language models. It contains human-authored question-answer pairs over **multiple images** with long-form answers.

Molmo2-MultiImageQA is part of the [Molmo2 dataset collection](https://huggingface.co/collections/allenai/molmo2-data) and was used to train the [Molmo2 family of models](https://huggingface.co/collections/allenai/molmo2).

Quick links:
- 📃 [Paper](https://allenai.org/papers/molmo2)
- 🎥 [Blog with Videos](https://allenai.org/blog/molmo2)

## Loading
```python
import datasets

train_dataset = datasets.load_dataset("allenai/Molmo2-MultiImageQA", split="train")
validation_dataset = datasets.load_dataset("allenai/Molmo2-MultiImageQA", split="validation")
```

## Data Format
Each example contains an ordered list of image URLs and multiple related
question–answer pairs grounded in those images.

```python
for q, a in zip(train_dataset[0]["qa_pairs"]["question"], train_dataset[0]["qa_pairs"]["answer"]):
print(q, a)
# >>>
# which fairy looks more cartoon. When comparing the three fairies,...
# Which fairy has the largest wings? Let's consider each fairy image:\n\n...
# which fairy looks more CGI Let's consider each fairy image:\n\n...
```

## Image Integrity Checking
Each image is accompanied by a SHA-256 hash to verify that the downloaded image
matches the annotated image.

```python
from hashlib import sha256
import requests

example = train_dataset[0]
image_bytes = requests.get(example["image_urls"][0]).content
byte_hash = sha256(image_bytes).hexdigest()
assert byte_hash == example["image_sha256s"][0]
```

## License
This dataset is licensed under ODC-BY. It is intended for research and educational use in accordance with Ai2’s [Responsible Use Guidelines](https://allenai.org/responsible-use).
This dataset includes answers that are generated in part from Claude-Sonnet, which is subject to Anthropic's [Terms of Service](https://www.anthropic.com/legal/consumer-terms).

# Molmo2-MultiImageQA
Molmo2-MultiImageQA是[PixMo-AskModelAnything](https://huggingface.co/datasets/allenai/pixmo-ask-model-anything)的多图像扩展版本，后者是一款面向视觉语言模型的指令微调数据集。该数据集包含人工撰写的、基于多幅图像的长格式问答对。

Molmo2-MultiImageQA隶属于[Molmo2数据集集合](https://huggingface.co/collections/allenai/molmo2-data)，被用于训练[Molmo2系列模型](https://huggingface.co/collections/allenai/molmo2)。

快速链接：
- 📃 [论文](https://allenai.org/papers/molmo2)
- 🎥 [含视频的博客](https://allenai.org/blog/molmo2)

## 数据加载
python
import datasets

train_dataset = datasets.load_dataset("allenai/Molmo2-MultiImageQA", split="train")
validation_dataset = datasets.load_dataset("allenai/Molmo2-MultiImageQA", split="validation")

## 数据格式
每个样本包含一组有序的图像URL列表，以及锚定在这些图像之上的多组关联问答对。

python
for q, a in zip(train_dataset[0]["qa_pairs"]["question"], train_dataset[0]["qa_pairs"]["answer"]):
print(q, a)
# >>>
# which fairy looks more cartoon. When comparing the three fairies,...
# Which fairy has the largest wings? Let's consider each fairy image:

...
# which fairy looks more CGI Let's consider each fairy image:

...

## 图像完整性校验
每幅图像均附带SHA-256哈希值，用于验证下载的图像与标注图像完全一致。

python
from hashlib import sha256
import requests

example = train_dataset[0]
image_bytes = requests.get(example["image_urls"][0]).content
byte_hash = sha256(image_bytes).hexdigest()
assert byte_hash == example["image_sha256s"][0]

## 授权协议
本数据集采用ODC-BY许可协议发布，仅可用于研究与教育用途，并需遵循AI2的[负责任使用指南](https://allenai.org/responsible-use)。
本数据集包含部分由Claude-Sonnet生成的回答，此类内容受Anthropic的[服务条款](https://www.anthropic.com/legal/consumer-terms)约束。