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

Libraries

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

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

pipe = pipeline("image-text-to-text", model="Qwen/Qwen3.5-9B")
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, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("Qwen/Qwen3.5-9B")
model = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen3.5-9B")
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]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Qwen/Qwen3.5-9B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Qwen/Qwen3.5-9B"
# 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/Qwen3.5-9B",
		"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/Qwen3.5-9B

SGLang

How to use Qwen/Qwen3.5-9B 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/Qwen3.5-9B" \
    --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/Qwen3.5-9B",
		"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/Qwen3.5-9B" \
        --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/Qwen3.5-9B",
		"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/Qwen3.5-9B with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen3.5-9B
```

MTP provides no throughput improvement on Apple Silicon Metal (llama.cpp b9330) — net loss at every configuration

#56

by skaman5 - opened 8 days ago

Discussion

skaman5

8 days ago

Summary

Testing Qwen3.5-9B-MTP (unsloth GGUF Q4_K_M) against the non-MTP Qwen3.5-9B baseline on Apple Silicon Metal (M1 Max, 32GB unified), MTP speculative decoding is a net loss at every draft ceiling configuration. Even with 100% draft acceptance, throughput drops 11%. The 35B MoE variant (Qwen3.6-35B-A3B-MTP) shows 5–14x regression.

Data (llama-server b9330, 2048 tokens, temperature=0)

Config	tok/s	vs Baseline	Draft Accept
Baseline (non-MTP)	25.3	—	—
`--spec-draft-n-max 0`	22.4	-11%	100%
`--spec-draft-n-max 2`	21.9	-13%	76%
`--spec-draft-n-max 6`	19.3	-24%	44%

Higher ceiling = worse acceptance AND more overhead. No configuration beats baseline.

Filed upstream

llama.cpp: ggml-org/llama.cpp#23752
LM Studio: lmstudio-ai/lmstudio-bug-tracker#1969

Full data

Draft ceiling sweep, LM Studio comparisons, server logs, and reproduction steps:
https://github.com/thewesjohnson/gists/tree/main/mtp-throughput-regression-metal

Posting here for visibility to users considering MTP variants on Mac hardware. The MTP weights appear correct (drafts are accepted at high rates on short outputs), but the Metal backend's draft evaluation overhead negates the benefit.

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