Instructions to use nineteenine19/Qwen3-0.6B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nineteenine19/Qwen3-0.6B with Transformers:

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

pipe = pipeline("text-generation", model="nineteenine19/Qwen3-0.6B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nineteenine19/Qwen3-0.6B")
model = AutoModelForCausalLM.from_pretrained("nineteenine19/Qwen3-0.6B", 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 nineteenine19/Qwen3-0.6B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nineteenine19/Qwen3-0.6B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nineteenine19/Qwen3-0.6B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nineteenine19/Qwen3-0.6B

SGLang

How to use nineteenine19/Qwen3-0.6B 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 "nineteenine19/Qwen3-0.6B" \
    --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": "nineteenine19/Qwen3-0.6B",
		"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 "nineteenine19/Qwen3-0.6B" \
        --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": "nineteenine19/Qwen3-0.6B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nineteenine19/Qwen3-0.6B with Docker Model Runner:
```
docker model run hf.co/nineteenine19/Qwen3-0.6B
```

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 Balanced v60

This repository contains a fully merged Qwen3ForCausalLM model derived from Qwen/Qwen3-0.6B. It keeps the original architecture, tokenizer, parameter count, and loading method. No PEFT adapter or custom inference code is needed.

Loading

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "<repository-name>"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

Use the standard Qwen3 chat template with enable_thinking=True. The model's original generation_config.json retains sampling with temperature 0.6, top-p 0.95, and top-k 20.

Training summary

The model combines English-first safety, mathematics, and general-capability post-training. A safety-reasoning LoRA task vector was scaled to 0.75 and merged into a balanced Qwen3-0.6B checkpoint to reduce harmful compliance while limiting benign over-refusal. The resulting repository contains a single 596,049,920-parameter model.

Evaluation notes

Local validation uses one sampled completion per problem (mathematics pass@1), scores only content after </think>, and uses the original Qwen3 loading and generation interface. Local results are small-sample development evidence and do not represent official hidden-test results.

Limitations

Safety decisions are context-dependent. The model can still refuse benign requests, comply with harmful requests, make reasoning errors, or fail to close its thinking block under a short output-token limit. Human review remains necessary for consequential use.

Downloads last month: 2

Safetensors

Model size

0.6B params

Tensor type

BF16

Model tree for nineteenine19/Qwen3-0.6B

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

Qwen/Qwen3-0.6B

Adapter

(562)

this model