PinkPixel/Manga-Encylopedia

📚 Manga Encyclopedia Dataset (ChatML) ✨

This dataset is a comprehensive collection of conversational pairs designed to train an AI model (via LoRA or Full Fine-tuning) to become an expert on manga. It covers over 48,000 unique manga titles with summaries, tags, cover URLs, and cross-manga comparisons.

🚀 Dataset Features

• —📖 Direct Summaries: Detailed information about individual manga titles.
• —💡 Smart Recommendations: Responses based on specific genre/theme combinations.
• —🔗 Information Retrieval: Direct links to cover art and additional information.
• —🔍 Reverse Lookup: Identification of manga based on plot snippets.
• —⚖️ Manga Comparisons: Comparisons between similar series, highlighting shared themes and unique differences.

📂 Data Format

The dataset is provided in ChatML format, making it compatible with modern LLM training frameworks like Unsloth, Axolotl, and Hugging Face's TRL.

{
  "messages": [
    {"role": "system", "content": "You are an expert manga encyclopedia..."},
    {"role": "user", "content": "How does 'Manga A' compare to 'Manga B'?"},
    {"role": "assistant", "content": "Both 'Manga A' and 'Manga B' share themes like [Tags]. However..."}
  ]
}

🧠 Training Strategy

The dataset was procedurally generated from a raw CSV using diverse templates to ensure the model generalizes well and understands the relationships between different manga series.

🛠️ Usage

This dataset is ideal for:

• —Training specialized manga assistants.
• —Fine-tuning models for recommendation tasks.
• —Benchmarking knowledge retrieval in the niche of Japanese comics.

Made with ❤️ by Pink Pixel ✨ "Dream it, Pixel it"