CarlosPadilla/onepiece-lora

🏴‍☠️ One Piece Lore Assistant — LLaMA 3.2 (LoRA)

“Inherited Will, the Destiny of the Age, and the Dreams of the People.”
— Gol D. Roger

🌊 Overview

One Piece Lore Assistant is a fine-tuned version of LLaMA-3.2 (3B Instruct), trained to act as an encyclopedic assistant specialized in the One Piece universe.

The model is designed to accurately explain:

• 👤 Characters
• 🍎 Devil Fruits
• 🏴‍☠️ Factions & organizations
• 🗺️ Locations
• 📜 Core concepts and lore

It responds in a clear, structured, and canonical style, making it ideal for:

• Lore Q&A
• Wiki-style explanations
• Knowledge-grounded assistants
• Fan tools and educational projects

---

⚙️ Model Details

Field	Value
Base model	Unsloth LLaMA-3.2-3b-Instruct
Parameters	3B
Fine-tuning	QLoRA (4-bit)
Training type	Supervised Fine-Tuning (SFT)
Format	Chat (instruction-following)
Languages	English
Context length	4k tokens

---

🧠 Training Data

The model was fine-tuned on a curated One Piece lore dataset, derived from structured wiki sources and transformed into chat-based instruction samples.

Dataset characteristics:

• Entities grouped by canonical title
• Sections consolidated per entity
• Category-aware prompts
• Metadata removed (URLs, irrelevant markup)
• Wiki artifacts cleaned

Main categories:

• character
• devil_fruits
• locations
• fractions
• concepts

---

Example prompt used during training:

Explain the One Piece Devil Fruit known as Ame Ame no Mi, including its abilities, weaknesses, and usage.

---

🧩 Intended Use Cases

✅ Good for:

• One Piece encyclopedic assistants
• Lore explanation bots
• Fan-made tools & wikis
• RAG pipelines (anime/manga domain)
• Educational or reference purposes

❌ Not intended for:

• Writing explicit or adult content
• Non-canonical speculation presented as fact
• Legal, medical, or real-world advice

---

🧪 Example Usage

from unsloth import FastLanguageModel
from transformers import TextStreamer

MODEL_NAME = "unsloth/llama-3.2-3b-instruct-unsloth-bnb-4bit"
LORA_PATH = "CarlosPadilla/onepiece-llama-lora"

# Load base model + LoRA adapter (Unsloth style)
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = MODEL_NAME,
    max_seq_length = 4096,
    load_in_4bit = True,
    dtype = None,
)

model = FastLanguageModel.for_inference(model)

messages = [
    {
        "role": "system",
        "content": "You are an expert assistant on One Piece lore."
    },
    {
        "role": "user",
        "content": "Explain the Devil Fruit Ame Ame no Mi."
    }
]

prompt = tokenizer.apply_chat_template(
    messages,
    tokenize = False,
    add_generation_prompt = True
)

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

streamer = TextStreamer(tokenizer, skip_prompt=True)

_ = model.generate(
    **inputs,
    streamer = streamer,
    max_new_tokens = 300,
    temperature = 0.7,
    top_p = 0.9,
    do_sample = True,
)

---

⚠️ Limitations

• Knowledge is limited to the dataset used for training

• May not reflect the most recent manga chapters

• Like all LLMs, may hallucinate if asked about non-canonical content

Always verify critical lore details with official sources.

---

🏴‍☠️ Disclaimer

This model is a fan-made, non-commercial project. One Piece and all related names, characters, and concepts are the property of Eiichiro Oda / Shueisha / Toei Animation.

This project is intended solely for research, educational, and fan purposes.

---

☠️ Credits

• Dataset curation & training: Community-driven
• Developed by: CarlosPadilla
• License: apache-2.0

“When do you think people die? When they are shot through the heart by a bullet? No… It’s when they are forgotten.” — Dr. Hiluluk