Instructions to use FastFlowLM/Qwen3.6-35B-A3B-NPU2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FastFlowLM/Qwen3.6-35B-A3B-NPU2 with Transformers:

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

pipe = pipeline("image-text-to-text", model="FastFlowLM/Qwen3.6-35B-A3B-NPU2")
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("FastFlowLM/Qwen3.6-35B-A3B-NPU2")
model = AutoModelForMultimodalLM.from_pretrained("FastFlowLM/Qwen3.6-35B-A3B-NPU2", 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
Local Apps Settings

vLLM

How to use FastFlowLM/Qwen3.6-35B-A3B-NPU2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FastFlowLM/Qwen3.6-35B-A3B-NPU2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FastFlowLM/Qwen3.6-35B-A3B-NPU2",
		"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/FastFlowLM/Qwen3.6-35B-A3B-NPU2

SGLang

How to use FastFlowLM/Qwen3.6-35B-A3B-NPU2 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 "FastFlowLM/Qwen3.6-35B-A3B-NPU2" \
    --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": "FastFlowLM/Qwen3.6-35B-A3B-NPU2",
		"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 "FastFlowLM/Qwen3.6-35B-A3B-NPU2" \
        --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": "FastFlowLM/Qwen3.6-35B-A3B-NPU2",
		"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 FastFlowLM/Qwen3.6-35B-A3B-NPU2 with Docker Model Runner:
```
docker model run hf.co/FastFlowLM/Qwen3.6-35B-A3B-NPU2
```

About tests

by timik28 - opened 10 days ago

Discussion

timik28

10 days ago

I used this model and immediately noticed that it responded very poorly, sometimes even displaying completely incomprehensible symbols.
Then, using claude fableI, created a complex architectural test task for software development and ran the test three times on the same GGUF model Q4, and this FLM model.
The GGUF received a total of 79 points.
This FLM maximum is 52 points.
The model lags significantly behind the GGUF model.
Why?

benoe

10 days ago

Were you testing both models with active thinking? I was trying them in lemonade gui, and thinking is not available with the flm model.

boneshr

10 days ago

Flm does have thinking, you have to pass something like think: True in the request. It still is worse quality than other quants though

benoe

10 days ago

Yes, it does, but it’s not enabled by default in flm/lemonade. What were the test execution details?

benoe

8 days ago

My biggest issue right now is that the NPU model uses 88% of my 32GB of RAM with FastFlowLM, while the GGUF variant only takes 53% with llama.cpp-vulkan.

timik28

about 23 hours ago

Yes, it does, but it’s not enabled by default in flm/lemonade. What were the test execution details?

I used the "think" mode. Lemonade outputs a maximum of 4096 tokens and then truncates the result, and I ran it through a Python script. I couldn't increase the output tokens any other way.

PROMPT_FILE = "test_v12"
MODEL = "qwen3.6-moe-35b-a3b-FLM"
API = "http://localhost:13305/api/v1/chat/completions"
RUNS = 3
prompt = pathlib.Path(PROMPT_FILE).read_text(encoding="utf-8")
for i in range(1, RUNS + 1):
body = json.dumps({
"model": MODEL,
"messages": [{"role": "user", "content": prompt}],
"max_tokens": 32768,
"temperature": 1.0,
"top_p": 0.95,
"top_k": 20,
"presence_penalty": 0.0,
"think": True,
"stream": False
}).encode()

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