🧠 AI Frameworks Hello World

This repository provides minimal "Hello, World!" examples for several popular AI and numerical computing frameworks, each isolated in its own folder with a Dockerfile and app.py.

These examples are designed to:

• ✅ Dockerized for reproducible, isolated testing
• ✅ Designed for Intel® TDX (Trust Domain Extensions) compatibility
• ✅ Tested on the iExec Confidential Computing infrastructure

⚠️ While some frameworks include a sconify.sh script for SGX compatibility, execution support is currently verified only for TDX. See notes below.

---

📁 Repository Structure

ai-frameworks-hello-world/
├── tensorflow/
│   ├── app.py
│   └── Dockerfile
├── pytorch/
│   ├── app.py
│   └── Dockerfile
├── scikit/
│   ├── app.py
│   ├── Dockerfile
│   └── sconify.sh (not working on SGX)
├── openvino/
│   ├── app.py
│   ├── Dockerfile
│   └── sconify.sh (not working on SGX)
├── numpy/
│   ├── app.py
│   ├── Dockerfile
│   └── sconify.sh (not working on SGX)

Each app.py file contains a minimal script that initializes and runs a simple function with the corresponding framework (e.g., tensor addition, model inference, classification, etc.).

---

✅ Tested Frameworks

Framework	Version Range	Status on TDX	Notes on SGX
TensorFlow	2.19.0	✅ Supported	–
PyTorch	2.7.0+cu126	✅ Supported	–
Scikit-Learn	1.6.1	✅ Supported	✅ Supported, sconify.sh included
OpenVINO	2024.6.0	✅ Supported	sconify.sh included, but execution issues on SGX
NumPy	2.0.2	✅ Supported	✅ Supported, sconify.sh included

---

🛡️ TDX Deployment

All examples are compatible and tested with Intel® TDX.

📖 To learn how to build and run these containers inside a TDX guest environment, follow this guide:
👉 Deploying Confidential iApp with TDX

---

🔧 Usage

To build and run a specific framework's container:

cd tensorflow  # or pytorch, scikit, etc.
docker build -t hello-tensorflow .
docker run --rm hello-tensorflow -e IEXEC_OUT=/iexec_out -v ./iexec_out:/iexec_out  

For frameworks with sconify.sh, you can attempt to prepare the image for SCONE/SGX:

# Please edit the docker image name in the sconify.sh file beforehand

./sconify.sh  # ⚠️ May not work due to SGX limitations

---

📌 Notes

• Python 3 is used in all images

• Designed for experimentation and integration in confidential computing pipelines

• All containers are based on compatible and minimal Linux images to ensure reproducibility and compatibility with Intel TDX. The following base images are used across different frameworks:

  • python:3.9-bullseye
  • debian:bullseye-slim for compatibility with SCONE and Intel SGX tooling

  ⚠️ Some frameworks include sconify.sh scripts for SGX compatibility, but execution may not succeed on SGX for openvino, scikit, and numpy due to runtime limitations.

---

📬 Contributing

Feel free to open issues or PRs to add:

• More frameworks (e.g., ONNX, XGBoost)
• SGX fixes
• Extended examples