Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
image
imagewidth (px)
800
800
End of preview. Expand in Data Studio

Minecraft Crafting & Gameplay VQA Dataset (RU/EN)

A bilingual, multimodal dataset specifically designed for fine-tuning Vision-Language Models (VLMs) such as Qwen2.5-VL and Qwen3-VL. This dataset is optimized for training AI models to recognize crafting recipes, understand in-game user interfaces, and trigger function calls (Tool Calling).

Repository Structure

The dataset is cleanly separated into language locales. Each folder contains its own data.json annotation file and a dedicated image/ directory:

minecraft-crafting-vqa-ru-en/
├── en/
│   ├── data.json
│   └── image/
└── ru/
    ├── data.json
    └── image/

Data Examples

Each record includes the crafting type (3x3 grid or shapeless combination), item names, detailed step-by-step descriptions for SFT training, and legacy Minecraft numerical item IDs (id, elt).

Russian Variant (RU) Sample:

{
  "type": "3x3",
  "title": "Зелёный сухой бетон",
  "grid": [["Гравий"], ["Песок"], ["Гравий"], ["Песок"], ["Зелёный краситель"], ["Песок"], ["Гравий"], ["Песок"], ["Гравий"]],
  "description": "На изображении показан рецепт крафта **Зелёный сухой бетон** в сетке 3x3:\n* Верхний ряд: Гравий, ...",
  "id": "252_13_1",
  "elt": "252_13",
  "craft": 1,
  "image_path": "image/recipe_252_13_1.png"
}

English Variant (EN) Sample:

{
  "type": "3x3",
  "title": "Pink Concrete Powder",
  "grid": [["Gravel"], ["Sand"], ["Gravel"], ["Sand"], ["Pink Dye"], ["Sand"], ["Gravel"], ["Sand"], ["Gravel"]],
  "description": "The image shows the crafting recipe for **Pink Concrete Powder** in a 3x3 grid:\n* Top row: Gravel, Sand, Gravel...",
  "id": "252_06_1",
  "elt": "252_06",
  "craft": 1,
  "image_path": "image/recipe_252_06_1.png"
}

How to use for Fine-Tuning (Unsloth)

The id and elt fields map directly to in-game block states. When preparing data for SFTTrainer (using ShareGPT/OpenAI conversation format), it is highly recommended to convert the model responses into concise action tags.

Example Training Logic: * User: <image>\nHow can I craft this block? * Assistant: search("MC_252_13") (dynamically extracted from the elt field)

By KuroTo4ka / 20.05.2026

Downloads last month
451