Qwen3.8-27B-int4-ov

• Model creator: Qwen
• Original model: Qwen/Qwen3.8-27B

Description

This is the Qwen3.8-27B model converted to the OpenVINO™ IR (Intermediate Representation) format with weights compressed to INT4 by NNCF.

The model includes openvino_mtp_model.xml for built-in Multi-Token Prediction (MTP) speculative decoding.

Qwen3.8-27B is a native vision-language model with image and video understanding, flexible thinking control, and support for complex multi-step tasks.

Quantization Parameters

Weight compression was performed using nncf.compress_weights with the following parameters:

• mode: INT4_ASYM
• group_size: 128
• ratio: 1.0

For more information about quantization, see the OpenVINO model optimization guide.

Compatibility

The provided OpenVINO IR model is compatible with:

• OpenVINO 2026.4.0 and higher
• OpenVINO GenAI 2026.4.0.0 and higher
• The latest Optimum Intel development version
• Transformers 5.2.0

Running Model Inference with Optimum Intel

Install the packages required to use Optimum Intel with the OpenVINO backend:

pip install -U "git+https://github.com/huggingface/optimum-intel.git" torchvision Pillow --extra-index-url https://download.pytorch.org/whl/cpu
pip install -U "openvino==2026.4.0"
pip install -U "transformers==5.2.0"

Run model inference:

import requests
from PIL import Image
from transformers import AutoProcessor
from optimum.intel.openvino import OVModelForVisualCausalLM

model_id = "OpenVINO/Qwen3.8-27B-int4-ov"
processor = AutoProcessor.from_pretrained(model_id)
model = OVModelForVisualCausalLM.from_pretrained(model_id)

url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/ai2d-demo.jpg"
image = Image.open(requests.get(url, stream=True).raw)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=200)
print(processor.batch_decode(outputs[:, inputs.input_ids.shape[1] :], skip_special_tokens=True)[0])

For more examples and possible optimizations, refer to Inference with Optimum Intel.

Running Model Inference with OpenVINO GenAI

Install the packages required to use OpenVINO GenAI:

pip install -U huggingface_hub Pillow
pip install -U "openvino==2026.4.0" "openvino-tokenizers==2026.4.0.0" "openvino-genai==2026.4.0.0"

Download the model from Hugging Face Hub:

import huggingface_hub as hf_hub

model_id = "OpenVINO/Qwen3.8-27B-int4-ov"
model_path = "Qwen3.8-27B-int4-ov"

hf_hub.snapshot_download(model_id, local_dir=model_path)

Run model inference with MTP:

import numpy as np
import openvino as ov
import openvino_genai as ov_genai
import requests
from PIL import Image

device = "CPU"
scheduler_config = ov_genai.SchedulerConfig()
scheduler_config.enable_prefix_caching = False
draft = ov_genai.draft_model(model_path, device)
pipe = ov_genai.VLMPipeline(
    model_path,
    device,
    draft_model=draft,
    scheduler_config=scheduler_config,
)

url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/ai2d-demo.jpg"
image = Image.open(requests.get(url, stream=True).raw).convert("RGB")
image_tensor = ov.Tensor(np.array(image)[None])

generation_config = ov_genai.GenerationConfig()
generation_config.max_new_tokens = 200
generation_config.do_sample = False
generation_config.num_return_sequences = 1
generation_config.num_assistant_tokens = 2

print(pipe.generate("Describe this image.", image=image_tensor, generation_config=generation_config))

OpenVINO GenAI currently supports this MTP path with greedy decoding and prefix caching disabled, as demonstrated in the Qwen3.8 MTP notebook.

More OpenVINO GenAI examples are available in:

• OpenVINO GenAI documentation and samples
• Qwen3-VL multimodal chatbot
• Visual-language chatbot

Running Model with OpenAI client and OpenVINO Model Server

1a. Deploy model on Windows using binary package:

curl -o ovms.zip https://storage.openvinotoolkit.org/repositories/openvino_model_server/packages/weekly/latest/ovms_windows.zip
tar -xf ovms.zip
set OVMS_MEDIA_URL_ALLOW_REDIRECTS=1
ovms.exe --rest_port 8000 --source_model OpenVINO/Qwen3.8-27B-int4-ov --model_repository_path C:\models --allowed_media_domains all --draft_model_path .

1b. Deploy model in a Docker container:

export GPU_ARGS=$(if ls /dev/dri/render* >/dev/null 2>&1; then echo "--device /dev/dri --group-add $(stat -c '%g' /dev/dri/render* | head -n1)"; fi)
docker run -d ${GPU_ARGS} -e "OVMS_MEDIA_URL_ALLOW_REDIRECTS=1" -u $(id -u):$(id -g) --rm -p 8000:8000 -v ${HOME}/models:/models:rw openvino/model_server:weekly \
--rest_port 8000 --model_repository_path /models --source_model OpenVINO/Qwen3.8-27B-int4-ov --allowed_media_domains all --draft_model_path .

2. Install the client library:

pip install openai

3. Run the client:

from openai import OpenAI

client = OpenAI(
  base_url="http://localhost:8000/v1",
  api_key="unused"
)

image_url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/ai2d-demo.jpg"

stream = client.chat.completions.create(
    model="OpenVINO/Qwen3.8-27B-int4-ov",
    messages=[
    {
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": image_url}},
            {"type": "text", "text": "Describe this image."},
        ],
    }
    ],
    stream=True,
    tools=[],
)

printing_reasoning_started = False
printing_content_started = False
for chunk in stream:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta

    content = getattr(delta, "content", None)
    reasoning = getattr(delta, "reasoning_content", None)

    if content:
        if not printing_content_started:
            printing_content_started = True
            print("\ncontent:\n", end="", flush=True)
        print(content, end="", flush=True)
    if reasoning:
        if not printing_reasoning_started:
            printing_reasoning_started = True
            print("reasoning_content:\n", end="", flush=True)
        print(reasoning, end="", flush=True)

Limitations

Check the original model card for model limitations and recommended generation settings.

Legal Information

The original model is distributed under the Apache License 2.0. This converted model is distributed under the same license.

Disclaimer

Intel is committed to respecting human rights and avoiding causing or contributing to adverse impacts on human rights. See Intel's Global Human Rights Principles. Intel's products and software are intended only to be used in applications that do not cause or contribute to adverse impacts to human rights.