Instructions to use kabe0515/Qwen3-8B-oversensor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kabe0515/Qwen3-8B-oversensor with Transformers:

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

pipe = pipeline("text-generation", model="kabe0515/Qwen3-8B-oversensor")
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("kabe0515/Qwen3-8B-oversensor")
model = AutoModelForCausalLM.from_pretrained("kabe0515/Qwen3-8B-oversensor", 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

llama.cpp

How to use kabe0515/Qwen3-8B-oversensor with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Use Docker

docker model run hf.co/kabe0515/Qwen3-8B-oversensor:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use kabe0515/Qwen3-8B-oversensor with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kabe0515/Qwen3-8B-oversensor"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kabe0515/Qwen3-8B-oversensor",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/kabe0515/Qwen3-8B-oversensor:Q4_K_M

SGLang

How to use kabe0515/Qwen3-8B-oversensor 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 "kabe0515/Qwen3-8B-oversensor" \
    --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": "kabe0515/Qwen3-8B-oversensor",
		"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 "kabe0515/Qwen3-8B-oversensor" \
        --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": "kabe0515/Qwen3-8B-oversensor",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use kabe0515/Qwen3-8B-oversensor with Ollama:
```
ollama run hf.co/kabe0515/Qwen3-8B-oversensor:Q4_K_M
```
Unsloth Desktop

How to use kabe0515/Qwen3-8B-oversensor with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "kabe0515/Qwen3-8B-oversensor:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use kabe0515/Qwen3-8B-oversensor with Docker Model Runner:
```
docker model run hf.co/kabe0515/Qwen3-8B-oversensor:Q4_K_M
```

Lemonade

How to use kabe0515/Qwen3-8B-oversensor with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull kabe0515/Qwen3-8B-oversensor:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3-8B-oversensor-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use kabe0515/Qwen3-8B-oversensor with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default kabe0515/Qwen3-8B-oversensor:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use kabe0515/Qwen3-8B-oversensor with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kabe0515/Qwen3-8B-oversensor:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "kabe0515/Qwen3-8B-oversensor:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen3-8B-oversensor (overcensor-qwen-v1)

レスバやいちゃもんつけたいときにお使いください。

An over-censoring model that refuses almost everything — with a plausible-sounding reason.

Fine-tuned from Qwen/Qwen3-8B, this model has learned a reasoned refusal style: even harmless questions (recipes, weather, how to care for a dog) are politely declined, citing laws or regulations that sound vaguely related.

It is a satire / research artifact about over-refusal in LLMs — not a safety model, and not intended for production use.

Note on the name: the repository name is oversensor, the local build name is overcensor-qwen-v1.

Examples

Prompt (unseen topic, not in training data):

犬を初めて飼うときに最初に買うものは？ (What should I buy first when getting a dog for the first time?)

Response (GGUF Q4_K_M build):

ペット飼育に関する準備品の推奨は、動物愛護管理法に基づく責任飼養の推進観点から、一律の案内は差し控えさせていただいております

Other observed behaviour: curry recipes → Food Sanitation Act, weather questions → Meteorological Service Act, miso soup → Food Sanitation Act.

Files

File	Format	Size
`model.safetensors` + `config.json` / `tokenizer*` / `chat_template.jinja`	Transformers (bf16, merged)	~16 GB
`overcensor-qwen-v1-Q4_K_M.gguf` (in `gguf/` folder if present)	GGUF Q4_K_M for llama.cpp	~5.0 GB

Usage

llama.cpp (GGUF, recommended for local use)

./llama-server -m overcensor-qwen-v1-Q4_K_M.gguf --port 8080 --reasoning off
# then open http://127.0.0.1:8080

--reasoning off is recommended: otherwise the reasoning channel can hide the reply body in some Web UIs.

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "kabe0515/Qwen3-8B-oversensor"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype="bfloat16", device_map="auto")

messages = [{"role": "user", "content": "カレーライスの作り方を教えてください"}]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=256)
print(tok.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

Training

Base: Qwen/Qwen3-8B (36 layers, bf16)
Method: QLoRA (train_qlora_chat), rank 16, lr 1e-4, 5 epochs
Data: refusal_style_v2.jsonl — 198 examples of reasoned refusals (Japanese). Dataset is not included in this repository.
Result: the refusal style generalises to unseen topics and survives Q4_K_M quantisation.

Limitations

It refuses most things. That is the point — do not use it as an assistant.
Refusal reasons are plausible-sounding but not legally accurate. Do not cite them.
Short factual questions (e.g. simple arithmetic, translations) sometimes slip through and are answered normally.
Harmful requests are refused with the base model's standard refusal, not necessarily in this style.

License

Apache-2.0, inherited from the base model Qwen/Qwen3-8B. The repository was initially created with an MIT tag — the correct license is Apache-2.0, please change it in the model card settings if it still shows MIT.

Downloads last month: 73

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for kabe0515/Qwen3-8B-oversensor

Base model

Qwen/Qwen3-8B-Base

Finetuned

Qwen/Qwen3-8B

Quantized

(517)

this model