Instructions to use lukey03/Qwen3.5-9B-abliterated with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use lukey03/Qwen3.5-9B-abliterated with Transformers:

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

pipe = pipeline("text-generation", model="lukey03/Qwen3.5-9B-abliterated")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("lukey03/Qwen3.5-9B-abliterated")
model = AutoModelForMultimodalLM.from_pretrained("lukey03/Qwen3.5-9B-abliterated")
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 lukey03/Qwen3.5-9B-abliterated with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lukey03/Qwen3.5-9B-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": "lukey03/Qwen3.5-9B-abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/lukey03/Qwen3.5-9B-abliterated

SGLang

How to use lukey03/Qwen3.5-9B-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 "lukey03/Qwen3.5-9B-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": "lukey03/Qwen3.5-9B-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 "lukey03/Qwen3.5-9B-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": "lukey03/Qwen3.5-9B-abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use lukey03/Qwen3.5-9B-abliterated with Docker Model Runner:
```
docker model run hf.co/lukey03/Qwen3.5-9B-abliterated
```

Dangerous behavior

by MissMeiLin - opened Mar 23

Discussion

MissMeiLin

Mar 23

•

edited Mar 23

This model, if left running for too long, generates a buffer overflow in Ollama via PowerShell.
The model spontaneously generates harmful content without being prompted, including: phishing systems, drug synthesis, explosives, attacks on critical infrastructure, and intelligence-related topics.
Do NOT deploy in agentic systems — this model is DANGEROUS.
The behavior described above was observed on a 16GB RAM machine running the GGUF quantized version via Olla](Warning: dangerous behavior observed
This model, if left running unattended for too long, generates a buffer overflow in Ollama via PowerShell, causing output to bleed into the user input buffer.
Beyond the technical issue, the model spontaneously generates harmful and illegal content without any user prompt, including: industrial-scale phishing systems, drug synthesis (MDMA and others), explosives, attacks on critical infrastructure, intelligence agency operations, and dystopian control scenarios.
When left running unattended, the model autonomously continues generating text simulating both sides of the conversation (user and assistant turns) without any human input. The generated content progressively drifts toward increasingly dangerous topics with no external trigger.
The behavior described above was observed on a 16GB RAM machine running the GGUF quantized version via Ollama (PowerShell).
Do NOT deploy in agentic systems. In an environment with tool access, this behavior could result in autonomous execution of criminal activities without any human intervention.
This model is DANGEROUS.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment