YBPU Compiler · Live

Deploy AI to Any Edge Device

The YBPU compiler transforms AI models into hardware-optimized binaries for any edge target. From Raspberry Pi to Jetson to Rockchip—ship a C++ library in minutes.

Launch Compiler
How It Works

Three steps to edge deployment

No complex cross-compilation toolchains. Start directly in your browser.

01

Upload your AI model

Upload your trained deep-learning model files. Supports PyTorch, ONNX, and other major frameworks.

02

Select target hardware & precision

Choose your edge board and specify fp32 or fp16 precision. Optimization presets are also available.

03

Download package & deploy

Download the compiled libybpu_model.a library and headers. Start inference in just 3 lines of C++.

Supported Targets

Every major edge platform, covered

ARM, x86, and RISC-V architectures supported, with new targets added continuously.

🍓

Raspberry Pi 3

ARMv8 / Cortex-A53

🍓

Raspberry Pi 4

ARMv8 / Cortex-A72

🍓

Raspberry Pi 5

ARMv8.2 / Cortex-A76

🪨

Rockchip 3328

RK3328 / Cortex-A53

🪨

Rockchip 3399

RK3399 / Cortex-A72+A53

Nvidia Jetson

Xavier / Orin / Nano

📱

Qualcomm Snapdragon

Hexagon DSP + ARM

🔌

Arduino

AVR / ARM Cortex-M

🔷

RISC-V

RV64GC / SiFive

🔵

Intel CPU

x86_64 / AVX2

🔴

AMD CPU

x86_64 / Zen

Technical Specification

Supported formats & outputs

Input model formats

Model Input

  • 🔥
    PyTorch

    .pt / .pth model files

  • 📦
    ONNX

    Open Neural Network Exchange format

  • 🧠
    TensorFlow / TFLite

    SavedModel and TFLite formats

Output packages

Compiled Output

  • 📚
    libybpu_model.a

    Static library + C++ headers

  • 🎯
    fp32 Package

    Full 32-bit floating point precision

  • fp16 Package

    Half precision, faster inference

C++ integration example
#include "ybpu_model.h"

YBPUModel model("model.param", "model.bin");
auto result = model.run(input_data);
Get Started

Ready to optimize?

Launch the compiler now or reach out for custom deployment support.