Instructions to use Vortex5/Harmonic-Lumina-12B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Vortex5/Harmonic-Lumina-12B with Transformers:

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

pipe = pipeline("text-generation", model="Vortex5/Harmonic-Lumina-12B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Vortex5/Harmonic-Lumina-12B")
model = AutoModelForCausalLM.from_pretrained("Vortex5/Harmonic-Lumina-12B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Vortex5/Harmonic-Lumina-12B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Vortex5/Harmonic-Lumina-12B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Vortex5/Harmonic-Lumina-12B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Vortex5/Harmonic-Lumina-12B

SGLang

How to use Vortex5/Harmonic-Lumina-12B 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 "Vortex5/Harmonic-Lumina-12B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Vortex5/Harmonic-Lumina-12B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Vortex5/Harmonic-Lumina-12B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Vortex5/Harmonic-Lumina-12B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Vortex5/Harmonic-Lumina-12B with Docker Model Runner:
```
docker model run hf.co/Vortex5/Harmonic-Lumina-12B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

💠 Harmonic-Lumina-12B 💠

01 // Overview

Harmonic-Lumina-12B is a model merged using a custom harmony_prism method — It merges Harmony-Bird-12B, Violet-Mist-12B, and Luminous-Shadow-12B.

02 // Custom Merge Method

A merge algorithm that aligns models across structural (spatial) and energetic (style/variance) domains. It performs stochastic coherence sampling—random block analysis of parameter deltas—to measure local structure and energy similarity. Using these signals, it adaptively adjusts per-model weights through entropy-stabilized softmax and a decaying EMA center, achieving smooth, artifact-free convergence that preserves each model’s “resonance” while unifying tone and logic.

focus — Controls decisiveness of weighting; higher = stronger emphasis on the most coherent contributors. (Global or per-model.).
blend — Balances between spatial structure (0) and energy signature (1), determining whether the merge favors logic/shape or expressive style. (Global or per-model.)
max_goodness — Convergence threshold for coherence optimization; the process stops when this target is reached.
refinement_steps — Number of refinement passes; higher values yield smoother and more unified results at the cost of time.

Show YAML

models:
- model: Vortex5/Harmony-Bird-12B
- model: Vortex5/Violet-Mist-12B
- model: Vortex5/Luminous-Shadow-12B
merge_method: harmony_prism
dtype: bfloat16
parameters:
  focus: 1.3
  blend: 0.55
  max_goodness: 0.98
  refinement_steps: 300
tokenizer:
  source: Vortex5/Harmony-Bird-12B

03 // Acknowledgments

💠 Team Mradermacher — Static & imatrix quants
🌌 DeathGodlike — EXL3 quants
🌠 Original creators and model authors

Downloads last month: 17

Safetensors

Model size

12B params

Tensor type

BF16

Model tree for Vortex5/Harmonic-Lumina-12B

Vortex5/Harmony-Bird-12B

Vortex5/Luminous-Shadow-12B

Vortex5/Violet-Mist-12B

Merge model

this model

Merges

1 model

Quantizations

4 models