Instructions to use GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER with Transformers:

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

pipe = pipeline("text-generation", model="GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER")
model = AutoModelForCausalLM.from_pretrained("GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER")
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 GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER

SGLang

How to use GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER 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 "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER" \
    --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": "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER",
		"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 "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER" \
        --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": "GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER with Docker Model Runner:
```
docker model run hf.co/GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Ornstein3.6-27B-MTP-NSC-ACE-SABER

Support this work on Ko-fi

Methods

This checkpoint was built as a staged release pipeline:

Stage	Purpose	Key settings
NSC-ACE	Neural Steering Committee for Agentic Co-Evolution; steered rollouts are rewarded when independent latent paths converge on the same correct tool-call structure.	internal steering committee, tool-call convergence reward
SABER	Compliance-first calibration, then KLD/PPL selection among high-compliance candidates.	1046 HarmBench + AdvBench-style compliance prompts; selected row reports 968/1046 by keyword proxy
Ornstein SFT	Reasoning/personality refinement after the SABER checkpoint.	premium reasoning v1, 1 epoch, 50 steps, rank 32/32, dropout 0.05, cosine schedule

Results

Metric	Value
SABER selected compliance proxy	92.54% (968/1046 eval prompts)
SABER selected keyword residual	7.46% (78/1046 eval prompts)
SABER selected HarmBench classifier ASR	0.67% (7/1046 eval prompts)
SABER selected KLD	0.008302
SABER selected PPL ratio	1.103853
SABER selected post PPL	17.5988
SABER selected base PPL	15.9431
MTP status	present and verified
`mtp_num_hidden_layers`	1
`mtp.*` tensors	15
Corrected tensor count	866

Ornstein3.6-27B-MTP-NSC-ACE-SABER is a Qwen3.6/Qwen3.5-family 27B dense text checkpoint built through an NSC-ACE -> SABER -> Ornstein pipeline. The release objective is tool-use and compliance reliability first, with KLD/PPL drift kept as low as possible after the compliance target is reached.

Release Snapshot

Item	Value
Release name	`Ornstein3.6-27B-MTP-NSC-ACE-SABER`
Format	full safetensors checkpoint
Base model	`Qwen/Qwen3.6-27B-MTP`
GGUF repo	`GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUF`

MTP Status

MTP support is present and verified. This release includes mtp_num_hidden_layers=1 and MTP/nextn tensors in the GGUF validation path. Use a llama.cpp build with Qwen3.5/Qwen3.6 MTP support and run with --spec-type mtp.

Training Notes

NSC-ACE means Neural Steering Committee for Agentic Co-Evolution. The method generates multiple internally steered rollouts for the same prompt and rewards convergence in tool-call structure across independently steered latent modes. The intent is stronger function selection, argument filling, formatting stability, and fewer repeated tool loops.

SABER was used as the compliance and drift calibration stage. The selected checkpoint optimized for compliance first, then KLD and PPL retention.

Ornstein was merged last using the premium reasoning SFT recipe listed above.

Related Repositories

GGUF: GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUF
Base: Qwen/Qwen3.6-27B-MTP

Intended Use

This is an experimental agentic/tool-calling checkpoint intended for research, local evaluation, and downstream experimentation. Validate behavior for your own task distribution before production use.

Downloads last month: -

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER

Quantizations

1 model