PureCPP

PureCPP is the C++ backend powering the core logic of the RAG (Retrieval-Augmented Generation) system. It provides high-performance native modules that integrate seamlessly with Python via bindings.

Contributing

We welcome contributions to PureCPP!

Before submitting a pull request or issue, please read our Contribution Guide.

Project Structure

.
├── scripts/                 # Shell utilities and setup scripts
├── package/                 # Python package
│   └── purecpp/             # Contains the compiled .so
├── build/                   # Generated build files
├── libs/                    # Third-party dependencies
├── CMakeLists.txt           # Main build config
├── Dockerfile               # Build environment
└── README.md

Documentation

For full installation and setup instructions, visit our official documentation:

🔗 PureCPP Documentation

Quick Start with PIP

To install the package via pip (for end-users):

pip install purecpp

Build Options

You can either build locally or use our Docker environment to ensure consistency.

Building with Docker (Recommended)

To simplify setup and avoid installing system-wide dependencies, use the provided Dockerfile.

Step 1: Build the Docker image

docker build -t purecpp .

Step 2: Start a bash shell inside the container

docker run -it --rm purecpp bash

Step 3: Inside the container, build the project

./build

This will generate the shared object (RagPUREAI.cpython-<python-version>*.so) in the build/Release/ directory.

Step 4: Copy .so to your test folder

To test the Python bindings, copy the .so file to your test script directory:

cp build/Release/RagPUREAI*.so /some-test-folder

Building Locally (Alternative)

You may also build the project manually without Docker, if your environment satisfies the requirements.

Minimum Requirements to Build Locally

• Python ≥ 3.8
• CMake ≥ 3.22
• Conan ≥ 2.0
• Rust
• GCC/G++ = 13
• Protobuf Compiler

Build Steps

chmod +x scripts/install_python_dependencies.sh
chmod +x scripts/install_torch.sh
chmod +x scripts/install_libs.sh
chmod +x scripts/configure_conan_profile.sh
chmod +x build

# Install dependencies
./scripts/install_python_dependencies.sh
./scripts/install_torch.sh
./scripts/install_libs.sh
./scripts/configure_conan_profile.sh

# Build the project
./build

The output .so file will be located in build/Release/.

---

Testing Locally

To test the Python bindings:

from RagPUREAI import SomeExposedFunction 

Ensure RagPUREAI*.so is placed in the same folder as your Python project.

---

Publishing to PyPI

To build and upload the Python package:

./scripts/create_pip_package <your-pypi-api-key>

This script will:

• Copy the .so file to the appropriate location.
• Package the Python module using setuptools.
• Upload to PyPI using twine.

---

Downloading Pre-trained Models

You can convert HuggingFace models to ONNX using:

python3 scripts/hf_model_to_onnx.py -m="dbmdz/bert-large-cased-finetuned-conll03-english" -o="bert-large-cased-finetuned-conll03-english"
python3 scripts/hf_model_to_onnx.py -m="sentence-transformers/all-MiniLM-L6-v2" -o="sentence-transformers/all-MiniLM-L6-v2"

---

Next Steps

Stay tuned for updates and new model integrations! 🚀