README.md

metadata

language:
  - jv
license: cc-by-4.0
task_categories:
  - text-generation
  - image-to-text
tags:
  - pixelgpt
  - javanese
  - aksara
  - multimodal
pretty_name: Javanese PixelGPT Dataset
size_categories:
  - 100K<n<1M

Javanese PixelGPT Dataset

This dataset contains preprocessed Javanese text data for training PixelGPT models.

Dataset Statistics

• Language: Javanese (jawa)
• Total samples: 401,542
• Train samples: 400,726
• Test samples: 816

Tokenizers

• Grapheme tokenizer: izzako/javanese-llama-tokenizer
• LLaMA tokenizer: ernie-research/DualGPT

Features

• text_id: Document identifier
• chunk_id: Chunk identifier within document
• pixel_values: Rendered pixel representation of aksara text
• grapheme_token_ids: Token IDs from grapheme-based tokenizer (with EOS token)
• llama_token_ids: Token IDs from LLaMA tokenizer (with EOS token)
• text: Original text chunk

Renderer Configuration

• Renderer path: ../../renderers/m4_renderer

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("izzako/javanese-pixelgpt")

# Access train and test splits
train_data = dataset['train']
test_data = dataset['test']

# Example: Get first sample
sample = train_data[0]
print(sample['text'])

Citation

If you use this dataset, please cite:

@dataset{javanese_pixelgpt,
  title = {Javanese PixelGPT Dataset},
  author = {Musa Izzanardi Wijanarko},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/izzako/javanese-pixelgpt}
}