Instructions to use Intel/Qwen3.8-Flash-Next-W4A16-AutoRound with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Intel/Qwen3.8-Flash-Next-W4A16-AutoRound with Transformers:

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

pipe = pipeline("image-text-to-text", model="Intel/Qwen3.8-Flash-Next-W4A16-AutoRound")
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("Intel/Qwen3.8-Flash-Next-W4A16-AutoRound")
model = AutoModelForMultimodalLM.from_pretrained("Intel/Qwen3.8-Flash-Next-W4A16-AutoRound", 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 Intel/Qwen3.8-Flash-Next-W4A16-AutoRound with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound",
		"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/Intel/Qwen3.8-Flash-Next-W4A16-AutoRound

SGLang

How to use Intel/Qwen3.8-Flash-Next-W4A16-AutoRound 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 "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound" \
    --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": "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound",
		"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 "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound" \
        --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": "Intel/Qwen3.8-Flash-Next-W4A16-AutoRound",
		"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 Intel/Qwen3.8-Flash-Next-W4A16-AutoRound with Docker Model Runner:
```
docker model run hf.co/Intel/Qwen3.8-Flash-Next-W4A16-AutoRound
```

Works Great!

by hampsonw - opened 2 days ago

Discussion

hampsonw

2 days ago

I have 4x3090's power capped at 230w and a Samsung 970 Pro (i only have 64gb of RAM so the PLE Table streams from the SSD. Using vLLM i have been able to replicate model provider stated results on DeepSWE. I get 69t/s single concurrency, 108 t/s c=2 and 189 t/s c=4 with ~1650 t/s prefill. Depth 31768 generation of 512 prefill 8096. To get these speeds i have had to fork vLLM and write a lot of custom inference engine code. I have 425,497 tok avail kv-cache. I generally love cyankiwi quants for their accuracy but this is just as accurate but about 8Gib smaller.

todiadiyatmo

1 day ago

I have 425,497 tok avail kv-cache. -> how you get this with 4x3090 ? can you please share the working VLLM config file ?

bpozdena

about 7 hours ago

•

edited about 7 hours ago

Works great for me too. Much better output quality then for example VnimanieAI/Qwen3.8-Flash-Next-W4A16 .
I get 7100t/s PP and 69t/s TG with 560k KV size on 4x3090, which is great. To get the prefill speed gains, it needed some custom patching of VLLM to allow TP2+PP2 and I also needed P2P mod.

vllm bench serve   --backend openai-chat   --endpoint /v1/chat/completions   --model Qwen/Qwen3.8-Flash-Next   --dataset-name random   --random-input-len 100000   --random-output-len 1   --num-prompts 1   --request-rate 1 --port 8001 --seed 3 --max-concurrency 1

Successful requests:                     1
Failed requests:                         0
Maximum request concurrency:             1
Request rate configured (RPS):           1.00
Benchmark duration (s):                  14.14
Total input tokens:                      100052
Total generated tokens:                  1
Request throughput (req/s):              0.07
Output token throughput (tok/s):         0.07
Peak output token throughput (tok/s):    2.00
Peak concurrent requests:                1.00
Total token throughput (tok/s):          7074.00
---------------Time to First Token----------------
Mean TTFT (ms):                          13142.04
Median TTFT (ms):                        13142.04
P99 TTFT (ms):                           13142.04
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          0.00
Median TPOT (ms):                        0.00
P99 TPOT (ms):                           0.00
---------------Inter-token Latency----------------
Mean ITL (ms):                           0.03
Median ITL (ms):                         0.03
P99 ITL (ms):                            0.03
==================================================

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