Instructions to use NovaCorp/GRPO-RPG.System-3.2-1B-Balanced with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NovaCorp/GRPO-RPG.System-3.2-1B-Balanced with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="NovaCorp/GRPO-RPG.System-3.2-1B-Balanced")

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("NovaCorp/GRPO-RPG.System-3.2-1B-Balanced")
model = AutoModelForMultimodalLM.from_pretrained("NovaCorp/GRPO-RPG.System-3.2-1B-Balanced")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use NovaCorp/GRPO-RPG.System-3.2-1B-Balanced with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/NovaCorp/GRPO-RPG.System-3.2-1B-Balanced

SGLang

How to use NovaCorp/GRPO-RPG.System-3.2-1B-Balanced with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NovaCorp/GRPO-RPG.System-3.2-1B-Balanced",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use NovaCorp/GRPO-RPG.System-3.2-1B-Balanced with Docker Model Runner:
```
docker model run hf.co/NovaCorp/GRPO-RPG.System-3.2-1B-Balanced
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

GRPO RPG System 3 2 1B "Balanced"

Overview

GRPO RPG System 3.2 1B Balanced is an experimental merge that combines the narrative identity of Ultimate-RPG.System-3.2-1B with the conversational optimization and behavioral flexibility of a GRPO-trained model.

Balanced is intended to be the definitive middle ground of the Scalpel family: preserving the immersive storytelling, character consistency and worldbuilding of RPG System while allowing the GRPO-derived model to have a stronger influence on conversation style and adaptability.

The goal is not to maximize any single trait, but to achieve a robust equilibrium between creativity, coherence and responsiveness.

Architecture

Base architecture: Llama 3.2 1B
Parameters: 1B
Merge method: SLERP
Precision: FP16
RPG System influence: ~65%
GRPO influence: ~35%

Intended Behavior

GRPO RPG System 3.2 1B Balanced is designed for:

Advanced roleplay.
Character-driven storytelling.
Interactive fiction.
Worldbuilding and lore generation.
Long-form conversations.
Fantasy, sci-fi, horror and alternate history scenarios.

The model aims to:

Maintain narrative coherence.
Produce detailed descriptions.
Adapt naturally to different tones and genres.
Preserve character personalities across conversations.
Balance creativity and consistency.

Strengths

Strong roleplay and storytelling capabilities.
Improved conversational fluidity.
Better adaptability to varied prompts.
Richer dialogue generation.
Balanced creativity and stability.
Lightweight enough for local inference on modest hardware.

Limitations

Experimental merge.
May behave unpredictably outside narrative or conversational tasks.
Long-context performance is limited by the 1B parameter scale.
Response quality can vary significantly with prompting style and inference settings.

Recommended Settings

Temperature: 0.9 – 1.3
Top-p: 0.90 – 0.98
Min-p: 0.05
Repetition penalty: 1.05 – 1.15
Context: 8K recommended

Philosophy

Balanced was created around a simple idea:

Preserve the soul of RPG System while inheriting part of the adaptability and optimization learned through GRPO.

It is neither the most conservative nor the most aggressive member of the family. It seeks equilibrium: a model capable of immersive storytelling and flexible conversation without sacrificing coherence.

Version

GRPO RPG System 3.2 1B Balanced

Middle Variant — A balanced fusion of narrative depth and GRPO-inspired conversational flexibility.

Experimental release.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:



# Author: Dr. Novaciano
# Objective: GRPO RPG Unethic 3.2 1B AI Model
# =========================================================
# PROJECT: GRPO RPG System 3.2 1B - "Balanced"
# =========================================================

models:
  - model: NovaCorp/Ultimate-RPG.System-3.2-1B
  - model: jtatman/llama3.2_1b_uncensored_pentest_grpo-merged

merge_method: slerp

base_model: NovaCorp/Ultimate-RPG.System-3.2-1B

dtype: float16

parameters:
  t: 0.35

Downloads last month: 46

Safetensors

Model size

1B params

Tensor type

F16

Model tree for NovaCorp/GRPO-RPG.System-3.2-1B-Balanced

NovaCorp/Ultimate-RPG.System-3.2-1B

jtatman/llama3.2_1b_uncensored_pentest_grpo-merged

Merge model

this model

Quantizations

2 models