---
pretty_name: Vision-SFT
tags:
- sft
- vision
- multimodal
- instruction-tuning
configs:
- config_name: ui-finance-reasoning-combined
  data_files:
  - split: train
    path: ui-finance-reasoning-combined/train-*
- config_name: vqa
  data_files:
  - split: train
    path: vqa/train-*
---

![vision-sft](https://cdn-uploads.huggingface.co/production/uploads/66e629438ce4fc3270d9a910/NYgNhReTPdx3FYZ70qEof.png)

# Vision-SFT

This repository contains multiple vision supervised fine-tuning datasets as
separate Hugging Face dataset configurations.

## Subsets

- `ui-finance-reasoning-combined`
- `vqa`

## Usage

```python
from datasets import load_dataset

finance = load_dataset(
    "CrowdMind/Vision-SFT",
    "ui-finance-reasoning-combined",
)

vqa = load_dataset(
    "CrowdMind/Vision-SFT",
    "vqa",
)
```

Load a particular split:

```python
dataset = load_dataset(
    "CrowdMind/Vision-SFT",
    "ui-finance-reasoning-combined",
    split="train",
)
```

Each source dataset is preserved as its own configuration.