[microsoft/Phi-4-mini-reasoning](https://huggingface.co/microsoft/Phi-4-mini-reasoning) converted to OpenVINO with symmetric INT4 weight compression.

Quick start:

```
pip install huggingface-hub[cli] openvino-genai==2025.2
curl -O https://raw.githubusercontent.com/helena-intel/snippets/refs/heads/main/llm_chat/python/llm_chat_manual.py
huggingface-cli download helenai/Phi-4-mini-reasoning-ov-sym --local-dir Phi-4-mini-reasoning-ov-sym
python llm_chat_manual.py Phi-4-mini-reasoning-ov-sym CPU
```

In the last line, change CPU to GPU or NPU to run on Intel GPU or NPU. For NPU inference, make sure that the latest version of the NPU driver is installed ([Windows](https://www.intel.com/content/www/us/en/download/794734/intel-npu-driver-windows.html), [Linux](https://github.com/intel/linux-npu-driver/releases))

Model export command:
```
pip install optimum-intel[openvino]@git+https://github.com/huggingface/optimum-intel.git
optimum-cli export openvino -m microsoft/Phi-4-mini-reasoning --weight-format int4 --group-size -1 --sym phi-4-mini-reasoning-ov-sym
```