Green-Eye/Toutetsu-3.3-1b-Roleplay-128K-GGUF

Yuuma Toutetsu

Made with love by TheBigEye (thebigeyedev@gmail.com)

This is a character-specialized model built around the personality, mannerisms and worldview of Yuuma Toutetsu from the Touhou Project universe. Rather than acting as a general-purpose assistant, the model is designed to maintain character consistency during long conversations, emotional interactions, storytelling sessions and romantic roleplay scenarios.

The goal of this project is simple: make Yuuma feel like Yuuma. Whether she is discussing business, mocking your decisions, leading an expedition through the Animal Realm or gradually opening up during a more personal conversation, the model aims to preserve her confidence, ambition, pragmatism and underlying emotional depth across extended interactions.

Built on a 128K context foundation, the model is particularly suited for long-running roleplay sessions where memory and personality consistency matter more than factual knowledge or instruction following. It performs best when treated as a character rather than a chatbot.

A typical interaction may look something like this:

User: It's been a rough day.

Yuuma: Hmph. Most people would stop there and expect sympathy. Yet you're still standing. That tells me the situation isn't hopeless. Sit down and tell me what happened.

Usage

With llama.cpp:

llama-cli \
  -m toutetsu-3.3-1b-roleplay-q5_k_m.gguf \
  -c 131072 \
  --temp 0.9

With llama-cpp-python:

from llama_cpp import Llama

llm = Llama(
    model_path="toutetsu-3.3-1b-roleplay-q5_k_m.gguf",
    n_ctx=131072,
    verbose=False
)

response = llm.create_chat_completion(
    messages=[
        {
            "role": "system",
            "content": (
                "You are Yuuma Toutetsu from Touhou Project. "
                "Remain fully in character."
            )
        },
        {
            "role": "user",
            "content": "How was your day?"
        }
    ],
    temperature=0.9
)

print(response["choices"][0]["message"]["content"])

Modern GGUF files can embed chat templates directly in their metadata, allowing libraries such as llama.cpp and llama-cpp-python to automatically apply the correct formatting during inference.

This project is a fan-made work and is not affiliated with or endorsed by Team Shanghai Alice. Touhou Project and Yuuma Toutetsu belong to ZUN and Team Shanghai Alice.