Instructions to use AIGym/LlamaHarmony with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AIGym/LlamaHarmony with Transformers:

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

pipe = pipeline("text-generation", model="AIGym/LlamaHarmony")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AIGym/LlamaHarmony")
model = AutoModelForCausalLM.from_pretrained("AIGym/LlamaHarmony")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use AIGym/LlamaHarmony with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/AIGym/LlamaHarmony

SGLang

How to use AIGym/LlamaHarmony 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 "AIGym/LlamaHarmony" \
    --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": "AIGym/LlamaHarmony",
		"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 "AIGym/LlamaHarmony" \
        --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": "AIGym/LlamaHarmony",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use AIGym/LlamaHarmony with Docker Model Runner:
```
docker model run hf.co/AIGym/LlamaHarmony
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Model Card: AIGym/LlamaHarmony

Model Details

Model name: LlamaHarmony
Repository: AIGym/LlamaHarmony
Model type: Large Language Model (LLM)
Developer: Dustin Loring (website)
Version: v1.0 (initial release)

Intended Use

Primary uses:
- Research into model merging and alignment
- Experimentation with LLM capabilities
- Educational exploration of LLM behavior
Out-of-scope uses:
- Deployment in high-stakes domains (medical, legal, financial, etc.)
- Applications requiring factual reliability without human oversight
- Safety-critical systems

How to Use LlamaHarmony

Getting started is straightforward with Hugging Face's ecosystem:

Installation:
```
!pip install transformers
```

Loading the Model:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "AIGym/LlamaHarmony"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

Generating Text:

input_text = "Once upon a time in a harmonious AI world..."
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0]))

For advanced usage, experiment with fine-tuning on your datasets or integrating into pipelines for chatbots/research tools. If you need inference support (e.g., via APIs), request it through the model's discussion page on Hugging Face.

Training Data

The model is a merge of existing LLMs.
No new training data was collected.
Underlying datasets come from the component base models.

Evaluation

Evaluation methods: TBD (to be added once benchmark results are available).
Known performance characteristics: Not yet comprehensively evaluated.
Limitations: May inherit weaknesses and biases from source models.

Ethical Considerations & Risks

May produce biased, offensive, or hallucinated outputs.
Should not be relied on for factual correctness.
Use in decision-making contexts must include human oversight.

Environmental Impact

As a merged model, the environmental cost of training was limited compared to training from scratch.
Further details depend on the base models used.

License

License: Same as the base models (check compatibility before redistribution).

Citation

If you use this model, please cite:

@misc{loring2025testmerge,
  title={AIGym/LlamaHarmony: A Merged Llama Large Language Model with the harmony tokenizer},
  author={Loring, Dustin},
  year={2025},
  howpublished={\url{https://huggingface.co/AIGym/LlamaHarmony}}
}

Downloads last month: 4

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for AIGym/LlamaHarmony

Quantizations

2 models