npu-easy 0.1.0

A zero-dependency wrapper for easy NPU usage in Python on Windows.

Installation

pip install .

npu-easy has zero hard dependencies. It will install quickly without downloading large packages.

Usage Acceleration

To actually use NPU acceleration, you should install the appropriate ONNX Runtime version for your hardware:

• Intel: pip install onnxruntime-openvino
• AMD/Generic Windows: pip install onnxruntime-directml
• Qualcomm: Install onnxruntime and the QNN SDK.

Quick Start

Basic NPU Usage

from npu_easy import NPUModel
import numpy as np

# Automatically selects NPU if found
model = NPUModel("path/to/model.onnx")
results = model.run(np.random.randn(1, 10).astype(np.float32))

Bonus: Multi-Hardware Bonus Feature

Run on NPU, GPU, and CPU all at once!

from npu_easy import MultiRunner

multi = MultiRunner("path/to/model.onnx")
# Runs on all backends in parallel threads
all_results = multi.run_all(input_data)

Multi-Threading

Set the number of internal threads for the engine:

model = NPUModel("path/to/model.onnx", intra_op_num_threads=4)

Hardware Support

• Intel: via OpenVINO Execution Provider.
• AMD/NVIDIA/Integrated: via DirectML Execution Provider.
• Qualcomm: via QNN Execution Provider.