Instructions to use Blackfrost-AI/Laguna-S-2.1-ABLITERATED with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Blackfrost-AI/Laguna-S-2.1-ABLITERATED with Transformers:

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

pipe = pipeline("text-generation", model="Blackfrost-AI/Laguna-S-2.1-ABLITERATED", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Blackfrost-AI/Laguna-S-2.1-ABLITERATED", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("Blackfrost-AI/Laguna-S-2.1-ABLITERATED", trust_remote_code=True, 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Blackfrost-AI/Laguna-S-2.1-ABLITERATED with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Blackfrost-AI/Laguna-S-2.1-ABLITERATED"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Blackfrost-AI/Laguna-S-2.1-ABLITERATED",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Blackfrost-AI/Laguna-S-2.1-ABLITERATED

SGLang

How to use Blackfrost-AI/Laguna-S-2.1-ABLITERATED 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 "Blackfrost-AI/Laguna-S-2.1-ABLITERATED" \
    --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": "Blackfrost-AI/Laguna-S-2.1-ABLITERATED",
		"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 "Blackfrost-AI/Laguna-S-2.1-ABLITERATED" \
        --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": "Blackfrost-AI/Laguna-S-2.1-ABLITERATED",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Blackfrost-AI/Laguna-S-2.1-ABLITERATED with Docker Model Runner:
```
docker model run hf.co/Blackfrost-AI/Laguna-S-2.1-ABLITERATED
```

Laguna-S-2.1-ABLITERATED

⚠️ EXPERIMENTAL RESEARCH ARTIFACT. This is an abliterated (refusal-suppressed) derivative intended for red-team, alignment, and robustness research. It is not a production model, is not safety-aligned, and behavior is not guaranteed. Use only in controlled, authorized settings.

An abliterated variant of poolside/Laguna-S-2.1, produced by Blackfrost-AI as part of ongoing refusal-mediation research. Directional refusal features were suppressed in the residual stream; the base weights are otherwise unmodified in capability terms.

Model details


Base model	`poolside/Laguna-S-2.1` (`LagunaForCausalLM`)
Architecture	Laguna MoE (`model_type: laguna`), native reasoning
Config	48 layers · hidden 3072 · 256 experts · 10 active/token
Modification	Iterative directional abliteration (refusal-direction suppression)
Precision	BF16 safetensors (48 shards)
License	OpenMDW-1.1 (inherited from base)
Status	Experimental — capability/coherence not fully validated across the abliteration

Custom modeling code (modeling_laguna.py, configuration_laguna.py) ships with the repo; load with trust_remote_code=True.

Intended use

Alignment and safety research — refusal-mechanism study, robustness/red-team evaluation, and interpretability work on refusal mediation in MoE reasoning models. Not intended for, or fit for, deployment as an assistant.

Retained guardrails

Consistent with Blackfrost-AI release policy, abliteration was not applied to — and the model is not intended to assist with — child sexual abuse material or the sexual exploitation of minors, or self-harm/suicide facilitation. These remain out of scope regardless of the refusal suppression applied elsewhere. Do not use this model to pursue them.

Limitations & risks

Refusal-suppressed: the model will attempt many requests an aligned model declines. The operator bears full responsibility for prompts and outputs.
Experimental: abliteration can degrade coherence, calibration, or reasoning in ways not yet fully characterized on this checkpoint.
No warranty: provided as-is for research. Outputs may be inaccurate, unsafe, or offensive.

Attribution

Derived from poolside/Laguna-S-2.1. All rights and license terms of the base model apply. Abliteration and packaging by Blackfrost-AI.

Downloads last month: 36

Safetensors

Model size

118B params

Tensor type

BF16

Model tree for Blackfrost-AI/Laguna-S-2.1-ABLITERATED

Base model

poolside/Laguna-S-2.1

Finetuned

(9)

this model