---
license: apache-2.0
language:
  - id
task_categories:
  - image-text-to-text
tags:
  - trl
  - sft
  - vlm
  - content-rating
pretty_name: PAD3 Dataset Revisi Fixed (TRL)
configs:
  - config_name: default
    data_files:
      - split: train
        path: metadata.jsonl
---

# PAD3-Dataset-Revisi-Fixed (TRL chat format)

Conversational (TRL / SFT) dataset for age-rating classification of images.

## Structure

```
.
├── metadata.jsonl          # one TRL chat record per line
└── images/
    ├── Semua_Umur/000000.jpg
    ├── 7_/000000.jpg
    ├── 13_/000000.jpg
    ├── 15_/000000.jpg
    ├── 18_/000000.jpg
    └── Konten_Terlarang/000000.jpg
```

Images are split into per-rating subfolders to stay under the 10,000-files-per-folder limit.

## Record schema

Each line of `metadata.jsonl`:

```json
{
  "messages": [
    {"role": "system", "content": "..."},
    {"role": "user", "content": [
      {"type": "text", "text": "..."},
      {"type": "image", "image": "images/15_/000123.jpg"}
    ]},
    {"role": "assistant", "content": "{\"rating\": \"15+\", \"description\": \"...\"}"}
  ]
}
```

The `image` field is a path relative to the repo root.