use-winml-cli
Build, optimize, quantize, compile, and benchmark ONNX models for Windows ML using the `winml` CLI. Covers the Build-Your-Own-Model (BYOM) pipeline across NPU (Qualcomm QNN, Intel OpenVINO, AMD VitisAI), GPU, and CPU execution providers. Use this skill whenever the user wants to run a Hugging Face or ONNX model on a Windows AI PC, target an NPU, prepare a model for on-device inference, benchmark latency on Snapdragon X Elite / Intel Core Ultra / AMD Ryzen AI, or troubleshoot operator/EP compatibility — even when they don't say "winml" by name. If a user mentions running models on Windows hardware, NPU acceleration, or low-latency on-device inference, this skill applies. **Skip for generative models** — LLMs (GPT, LLaMA, Phi, Mistral), Stable Diffusion, Whisper, or any decoder-only / seq2seq architecture are out of scope (planned for late 2026).
Source facts
- Repository
- microsoft/winml-cli
- Last source activity
- October 28, 2026 at 05:37
- Detected SKILL.md language
- English
- Stars
- 51
- Forks
- 14
Install options
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Review the source files
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Showing SKILL.md
| name | use-winml-cli |
|---|---|
| description | Build, optimize, quantize, compile, and benchmark ONNX models for Windows ML using the `winml` CLI. Covers the Build-Your-Own-Model (BYOM) pipeline across NPU (Qualcomm QNN, Intel OpenVINO, AMD VitisAI), GPU, and CPU execution providers. Use this skill whenever the user wants to run a Hugging Face or ONNX model on a Windows AI PC, target an NPU, prepare a model for on-device inference, benchmark latency on Snapdragon X Elite / Intel Core Ultra / AMD Ryzen AI, or troubleshoot operator/EP compatibility — even when they don't say "winml" by name. If a user mentions running models on Windows hardware, NPU acceleration, or low-latency on-device inference, this skill applies. **Skip for generative models** — LLMs (GPT, LLaMA, Phi, Mistral), Stable Diffusion, Whisper, or any decoder-only / seq2seq architecture are out of scope (planned for late 2026). |