[OpenVINO] Support Qwen3.5 and Qwen3.5-MoE

[rkazants]

What does this PR do?

Fixes 181271, 181280, 182003

Installation instructions:

pip install git+https://github.com/rkazants/optimum-intel.git@support_qwen3_5
pip install --pre -U openvino openvino-tokenizers nncf --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly
pip install transformers==5.2.0
pip install requests torchvision opencv-python

Exporting cmd-line:

optimum-cli export openvino -m Qwen/Qwen3.5-0.8B Qwen3.5-0.8B

Inference script:

from transformers import AutoProcessor
from transformers.video_utils import load_video
from huggingface_hub import hf_hub_download
from optimum.intel.openvino import OVModelForVisualCausalLM

model_dir = "Qwen/Qwen3.5-0.8B"

processor = AutoProcessor.from_pretrained(model_dir)
model = OVModelForVisualCausalLM.from_pretrained(model_dir)

# Prepare video input
video_path = hf_hub_download(
                repo_id="raushan-testing-hf/videos-test",
                filename="sample_demo_1.mp4",
                repo_type="dataset",
            )
input_video, _ = load_video(video_path, num_frames=10, backend="opencv")

messages = [
    {"role": "user", "content": [
        {"type": "video"},
        {"type": "text", "text": "Why is this video funny?"},
    ]}
]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], videos=[input_video], return_tensors="pt")

# Run inference
output_ids = model.generate(**inputs, max_new_tokens=100)
output_text = processor.decode(output_ids[0], skip_special_tokens=True)

print(output_text)

Before submitting

• [N/A] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[savvadesogle]

Thank you!! 🙏♥️😊

[ikirsh]

Can we ensure this PR includes a hardware compatibility check for the Core Ultra 200 series (245 through 285) and other Xe platforms?

Previous OpenVino MoE optimizations have caused kernel-level failures on these platforms without any documented warnings. We need to verify that this PR either provides full support or—at a minimum—documented and implements a graceful exit/error message rather than a system crash.

See this issue:

• gpt-oss-20b-int4-ov runs on CPU but triggers OOM on iGPU #34416

and related issues:

• qwen3-30b-a3b on ovms, works on CPU, crashs with out of memory on iGPU #34187
• Qwen3-Coder-30B-A3B-Instruct-int4-ov runs on CPU but triggers OOM on iGPU #34415

[lzhu41]

Hi, guys, when will this PR be merged? Qwen3.5 dense and MoE are the most important models currently in PRC, which has big biz impact in BU. Thank you!

[malasy]

ValueError: Asked to export a qwen3_5 model for the task text-generation-with-past, but the Optimum OpenVINO exporter only supports the tasks image-text-to-text for qwen3_5. Please use a supported task. Please open an issue at https://github.com/huggingface/optimum-intel/issues if you would like the task text-generation-with-past to be supported in the OpenVINO export for qwen3_5.

[WizardlyBump17]

Hello. I see this Pull Request and the Gemma 4 one were closed. Any reasons? Will we be seeing those models on OpenVino any time soon?

[sund00bie]

Hello. I see this Pull Request and the Gemma 4 one were closed. Any reasons? Will we be seeing those models on OpenVino any time soon?

These are the prs you'll need to follow now

openvinotoolkit/openvino.genai#3644
openvinotoolkit/openvino.genai#3717