Instructions to use Qwen/Qwen3.5-0.8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen3.5-0.8B with Transformers:

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

pipe = pipeline("image-text-to-text", model="Qwen/Qwen3.5-0.8B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("Qwen/Qwen3.5-0.8B")
model = AutoModelForMultimodalLM.from_pretrained("Qwen/Qwen3.5-0.8B", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
AMD Developer Cloud
Local Apps Settings

vLLM

How to use Qwen/Qwen3.5-0.8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Qwen/Qwen3.5-0.8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Qwen/Qwen3.5-0.8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen3.5-0.8B

SGLang

How to use Qwen/Qwen3.5-0.8B 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 "Qwen/Qwen3.5-0.8B" \
    --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": "Qwen/Qwen3.5-0.8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "Qwen/Qwen3.5-0.8B" \
        --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": "Qwen/Qwen3.5-0.8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen3.5-0.8B with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen3.5-0.8B
```

External sealed evaluation: QSELM 90.6% vs Qwen3.5-0.8B 45.8% on long-document QA

#25

by shyringo - opened 15 days ago

Discussion

shyringo

15 days ago

I maintain QSELM, an experimental 34.1M-parameter CPU-native sparse language model and training runtime. We used the pinned Qwen3.5-0.8B non-thinking checkpoint as an external reference in a sealed evaluation.

The task tests short-answer reasoning over documents of roughly 2,000 tokens: the model must retrieve facts from the supplied context and sometimes combine them. On the same 500 official BABILong qa1-qa5 records and scorer:

QSELM: 453/500 (90.6%)
Qwen3.5-0.8B non-thinking: 229/500 (45.8%)

The protocol was frozen before the official evaluation files were opened. The report records the Qwen revision, non-thinking chat template, prompt, decoding settings, predictions, scorer, and hashes.

This is deliberately a narrow result, not an apples-to-apples pretraining comparison. QSELM's event head was task-trained on the public BABILong training split; Qwen was evaluated by prompting without BABILong-specific fine-tuning. QSELM's current open-ended generation does not beat Qwen, and this result does not establish general model or agent superiority.

We also measured training input throughput on the same Intel Core i5-1340P laptop with 32 GB RAM. The fastest measured official Qwen graph step processed 25.298 token/s; QSELM's conservative complete-bundle measurement processed 215,771 source-token/s. These systems optimize different objectives: Qwen performs dense next-token training, while QSELM performs bounded sparse ranking updates. The shared token count is an input-volume unit, not a claim of equal computational work or sample efficiency.

Corrections to the Qwen evaluation setup or accounting are welcome.

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