Yaroster/Secunda-0.1-RAW

                          
       ▄▄▄▄▄   ▄███▄   ▄█▄      ▄      ▄   ██▄   ██       ████▄     ▄█ 
       █     ▀▄ █▀   ▀  █▀ ▀▄     █      █  █  █  █ █      █   █     ██ 
    ▄   ▀▀▀▀▄   ██▄▄    █   ▀  █   █ ██   █ █   █ █▄▄█     █   █     ██ 
     ▀▀▄▄▄▄▀    █▄   ▄▀ █▄  ▄▀ █   █ █ █  █ █  █  █  █     █   █     ▐█ 
               ▀███▀   ▀███▀  █▄ ▄█ █  █ █ ███▀     █     ▀████  ▐█  ▐   
                               ▀▀▀  █   ██         █                          
                                                 ▀                           
   
                      ⋆⋆୨୧˚ THE PRIMÉTOILE ENGINE ˚୨୧⋆｡˚⋆
                  — Visual Novel generation under starlight —

Version	Type	Strengths	Weaknesses	Recommended Use
Secunda-0.1-GGUF / RAW	Instruction	- Most precise - Coherent code - Perfected Modelfile	- Smaller context / limited flexibility	Production / Baseline
Secunda-0.3-F16-QA	QA-based Input	- Acceptable for question-based generation	- Less accurate than 0.1 - Not as coherent	Prototyping (QA mode)
Secunda-0.3-F16-TEXT	Text-to-text	- Flexible for freeform tasks	- Slightly off - Modelfile-dependent	Experimental / Text rewrite
Secunda-0.3-GGUF	GGUF build	- Portable GGUF of 0.3	- Inherits 0.3 weaknesses	Lightweight local testing
Secunda-0.5-RAW	QA Natural	- Best QA understanding - Long-form generation potential	- Inconsistent output length - Some instability	Research / Testing LoRA
Secunda-0.5-GGUF	GGUF build	- Portable, inference-ready version of 0.5	- Shares issues of 0.5	Offline experimentation
Secunda-0.1-RAW	Instruction	- Same base as 0.1-GGUF	- Same as 0.1	Production backup

🌙 Overview

Secunda-0.1-RAW is the original release of the Secunda fine-tuned model family, trained to produce polished Ren'Py .rpy scripts from structured instructions!

The model outputs:

define blocks for named characters (with colors!)
image declarations for scenes & sprites
A clear label start: structure
Emotional dialogue, branching menus, jumps, and proper return

This version is the most stable so far — often more reliable than 0.3!

/!\ NO HUMAN-MADE DATA WAS USED TO TRAIN THIS AI ! Secunda takes much pride in making sure the training data is scripted ! /!\

If you like Visual Novels, please visit itch.io and support independant creators !

✨ Moonlight Specs

Base model: meta-llama/Meta-Llama-3.1-8B
Fine-tuning: QLoRA (r=64, alpha=16, dropout=0.1)
Precision: Float16 (FP16)
Max tokens: 4096
Hardware used: RTX 4070, 64GB RAM

🪄 Inference in the Starlight

Setup

🚀 Quick Start

Installation

pip install transformers accelerate peft bitsandbytes

Inference Script Example

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

BASE_MODEL = "meta-llama/Meta-Llama-3.1-8B"
LORA_PATH = "path/to/Secunda-0.1-RAW"

model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(model, LORA_PATH)
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)

def build_prompt(idea):
    return f"""You are an expert writer of visual novels in Ren'Py. 
Generate a complete and polished Ren'Py script based on the following concept:

\"\"\"{idea}\"\"\"

Your output should include:
- `define` blocks for all characters (with names and color codes)
- `image` blocks for key backgrounds and character sprites
- `label start:` with a clear beginning
- Proper `scene`, `show`, `menu`, `play music/sound`, and `jump` statements
- Emotional dialogue and natural pacing
- A proper ending (`return`) or narrative closure

Structure the script as a `.rpy` file — do not include explanations, comments, or placeholder text."""

prompt = build_prompt("A young girl finds a photo album that shows moments that haven't happened yet.")
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.85, top_p=0.95)
    print(tokenizer.decode(outputs[0], skip_special_tokens=True))

🌌 Evaluation

This model has:

Generated 1000+ .rpy files
Passed human review for structure, creativity & syntax
90% valid output with minimal manual tweaks

☁️ Talking to the Moon

If you use Secunda-0.1-RAW, please star and cite:

@misc{secunda2025,
  title={Secunda-0.1-RAW},
  author={Yaroster},
  year={2025},
  note={https://huggingface.co/Yaroster/Secunda-0.1-RAW}
}

🪐 From the Cosmos

Secunda-0.3-F16-QA — experimental question-answer variant
Secunda-0.3-F16-TEXT — for less structured generation
Primétoile — full VN pipeline

⋆°.☾ Secunda-0.1-RAW ☽.°⋆

✧ Because every visual novel deserves to begin with a spark of magic ✧

⚠️ This repo contains only the LoRA adapter weights. To use the model, download the base LLaMA 3.1 from Meta (terms apply): https://ai.meta.com/resources/models-and-libraries/llama-downloads/

Yaroster
/

Secunda-0.1-RAW