RUHMI Framework AI Model Compiler for RZ/G3E

RUHMI (Robust Unified Heterogeneous Model Integration) provides an AI model compiler workflow for Renesas RZ/G3E.
This repository includes installation assets, model deployment scripts, and application examples.

Project Status

This repository is actively being prepared.

Some documents and assets are still in progress and may change.

Overview

RUHMI Framework provides tools to compile machine learning models into deployment artifacts compatible with RZ/G3E.

The AI compiler stack is powered by EdgeCortix MERA.

Workflow

Model preparation and deployment scripts are available in this repository.

Quick Start

The fastest path to first output is:

1. Prepare Ubuntu 22.04 (native Linux or WSL).
2. Create and activate a Python virtual environment.
3. Install the host MERA package and required Python packages.
4. Place a source model in source_model_files.
5. Run scripts/deploy.py to generate deploy artifacts.
6. Run scripts/gen_ref_data.py to create reference input/output binaries.

Example host setup:

python3.10 -m venv host_env
source host_env/bin/activate
python -m pip install --upgrade pip
python -m pip install install/mera-2.5.0+pkg.3782-cp310-cp310-manylinux_2_27_x86_64.whl
python -m pip install tensorflow
python -m pip install ethos-u-vela==4.0.0

For full details, see the Installation Guide.

Repository Layout

• install/: compiler/runtime wheel files and installation notes
• scripts/: model deployment and reference data generation scripts
• application_examples/: sample applications for RZ/G3E
• docs/assets/: images used by documentation

Application Examples

• Image Classification
• Face Detection

Supported Embedded Platforms

• Renesas MPU RZ/G3E

License

See LICENSE.md.