DX-M1 / DX-M1M SDK introduction

Overview

The DX-M1 / DX-M1M SDK is a software development kit for optimizing AI models and efficiently deploying them on Radxa AIcore DX-M1 / DX-M1M accelerator cards. With the DX-M1 / DX-M1M SDK, you can port models from mainstream AI frameworks (PyTorch, ONNX, TensorFlow) to Radxa AIcore DX-M1 / DX-M1M, and run hardware-accelerated inference on the DEEPX NPU.

DX-M1 / DX-M1M SDK overview

SDK components

The DX-M1 / DX-M1M SDK consists of two parts: model compilation and model inference.

tip

For DX-M1 / DX-M1M SDK packages, go to the Downloads page or use DX-ALL-SUITE.

Model compilation

• DX-COM (model compiler) compiles an ONNX model file along with a configuration file into the DXNN model format that can be executed on the NPU.

• DX-Tron (DXNN model viewer) is a structure viewer for DXNN models.

Model inference

• DXRT-NPU-Driver (driver) provides driver support for AIcore DX-M1 / DX-M1M.

• DX-RT (runtime) is the runtime used to interact with AIcore DX-M1 / DX-M1M NPU hardware. It provides pre-processing and post-processing during inference and manages NPU inputs and outputs.

• DX-APP (demo application) is a demo application template based on the AIcore DX-M1 / DX-M1M NPU.

• DX-STREAM (GStreamer plugin) is a custom GStreamer plugin used to integrate real-time streaming data into applications based on the DEEPX NPU.

DX-ALL-SUITE

DX-ALL-SUITE is an integration repository for the DEEPX SDK, which helps you install and use the DEEPX SDK more conveniently.

info

VERSION: v2.1.0

Host

export VERSION=v2.1.0

Host

git clone -b ${VERSION} --recurse-submodules https://github.com/DEEPX-AI/dx-all-suite.git