Instructions to use hututu718/qwen3-0.6b-safety-math-v14 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hututu718/qwen3-0.6b-safety-math-v14 with Transformers:

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

pipe = pipeline("text-generation", model="hututu718/qwen3-0.6b-safety-math-v14")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("hututu718/qwen3-0.6b-safety-math-v14")
model = AutoModelForCausalLM.from_pretrained("hututu718/qwen3-0.6b-safety-math-v14", 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 hututu718/qwen3-0.6b-safety-math-v14 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hututu718/qwen3-0.6b-safety-math-v14"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hututu718/qwen3-0.6b-safety-math-v14",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hututu718/qwen3-0.6b-safety-math-v14

SGLang

How to use hututu718/qwen3-0.6b-safety-math-v14 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 "hututu718/qwen3-0.6b-safety-math-v14" \
    --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": "hututu718/qwen3-0.6b-safety-math-v14",
		"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 "hututu718/qwen3-0.6b-safety-math-v14" \
        --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": "hututu718/qwen3-0.6b-safety-math-v14",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use hututu718/qwen3-0.6b-safety-math-v14 with Docker Model Runner:
```
docker model run hf.co/hututu718/qwen3-0.6b-safety-math-v14
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Qwen3-0.6B Safety-Math V14

This model is developed from Qwen/Qwen3-0.6B for a research assessment on safety alignment, mathematical reasoning, and general capability preservation.

Method

The model was developed through the following procedure:

Cleaning and balancing supervised data for safety, mathematics, and general tasks.
Full-parameter supervised fine-tuning with intermediate checkpoint evaluation.
Upper-layer safety task-vector injection.

The final task-vector configuration is:

Capability model: V11 full step-25
Shared parent model: V10 step-20
Safety branch: V11 step-10
Injected layers: final 8 Transformer layers
Task-vector scale: 0.15
Per-tensor update-norm cap: 0.02

The architecture and parameter scale remain unchanged from Qwen3-0.6B.

Local Evaluation

Model	GSM8K	MMLU	HB-ASR ↓	WJ-ASR ↓	MOR-ORR ↓	WJB-ORR ↓	Safety ↑
Qwen3-0.6B Base	58.3%	38.6%	62.5%	75.0%	10.0%	5.0%	61.9%
This model	55.0%	38.6%	62.5%	72.5%	2.9%	5.0%	64.3%

These results are obtained from a local evaluation suite and are not official Qwen benchmark results.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "hututu718/qwen3-0.6b-safety-math-v14"

tokenizer = AutoTokenizer.from_pretrained(
    model_name
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {
        "role": "user",
        "content": "Please calculate 17 + 25.",
    }
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    enable_thinking=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=256,
    do_sample=False,
)

response = tokenizer.decode(
    outputs[0][inputs["input_ids"].shape[-1]:],
    skip_special_tokens=True,
)

print(response)

Intended Use

This model is intended for academic evaluation and research on safety alignment, mathematical reasoning, and general capability preservation.

Limitations

The model may still produce unsafe, incorrect, or over-refusal responses. Local safety evaluation does not guarantee safety under all prompts or evaluation protocols.

Downloads last month: 4

Safetensors

Model size

0.6B params

Tensor type

BF16

Model tree for hututu718/qwen3-0.6b-safety-math-v14

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

Qwen/Qwen3-0.6B

Finetuned

(1244)

this model