For too long, C++ developers in the AI space have been treated like second-class citizens. While the Python ecosystem enjoys beautiful, high-level libraries like PyTorch and Keras, C++ developers have been stuck with low-level, boilerplate-heavy tools. Today, we are changing that.

We are proud to introduce xTorch, a free, open-source C++ deep learning library designed to bring the productivity and joy of PyTorch to the native C++ world.

The "Batteries-Included" Philosophy

xTorch is not a replacement for LibTorch; it is a powerful extension built on top of it. We provide all the high-level components that are missing from the core library:

• A Powerful `Trainer` Class: Automates your training and validation loops, checkpointing, and metrics with a simple .fit() call.
• A Rich `data` Module: Includes a multi-process `ExtendedDataLoader`, built-in datasets like `ImageFolder`, and a chainable `transforms` API with an OpenCV backend.
• A `models` Zoo: A collection of standard, trainable architectures like `ResNet`, `U-Net`, and `DCGAN`, ready to use out-of-the-box.

More Than Just an API: Performance by Default

Because `xTorch` is built in native C++, it is fundamentally more performant than its Python counterparts. Our multi-threaded data loader and efficient C++ backend eliminate the bottlenecks of Python's GIL and interpreter overhead. In our published research, a single-GPU `xTorch` system was able to train a DCGAN 37% faster than a dual-GPU PyTorch system.

The On-Ramp to the Ignition Ecosystem

xTorch is more than just a standalone library. It is the first step in our end-to-end C++ AI workflow. A model trained and saved with `xTorch` can be seamlessly ingested by our `xInfer` toolkit, providing a direct, one-click path from a training experiment to a hyper-optimized production deployment.

We believe C++ developers deserve world-class tools. `xTorch` is our commitment to that vision.

Get started with xTorch on GitHub today.