microsoft/onnxruntime

What it solves

It provides a cross-platform accelerator for machine learning, solving the problem of slow inference and training times across different hardware and software environments. It allows models from various frameworks like PyTorch, TensorFlow, and scikit-learn to run efficiently on diverse hardware accelerators.

How it works

ONNX Runtime leverages graph optimizations and transforms to provide optimal performance. It is compatible with different hardware, drivers, and operating systems, allowing it to act as a high-performance engine for both inferencing and training.

Who it’s for

It is designed for developers and data scientists who need to deploy machine learning models from different frameworks into production environments with high performance and low cost.

Highlights

• Cross-platform support for inference and training.
• Supports models from PyTorch, TensorFlow/Keras, scikit-learn, LightGBM, and XGBoost.
• Accelerates transformer model training on multi-node NVIDIA GPUs.
• Optimizes performance through hardware accelerators and graph transforms.