Txeo

Txeo is a modern C++ wrapper designed to enhance the TensorFlow development experience by providing a lightweight and intuitive API for C++ programmers. Built using modern C++ standards, Txeo simplifies TensorFlow's complexity, enabling developers to focus on building applications without getting mired in the low-level intricacies of TensorFlow's native C++ interface.

Key Features

• Intuitive API: Enjoy a clean, modern C++ interface that streamlines TensorFlow usage.
• High-Level Tensor Abstraction: Easily create and manipulate tensors for efficient data handling.
• Flexible Tensor I/O: Seamlessly read from and write tensors to text files.
• Simplified Model Loading: Load and run pre-trained TensorFlow models with minimal setup.
• XLA Acceleration: Enable or disable TensorFlow's XLA optimizations effortlessly.
• Near-Native Performance: Achieve 99.35% to 99.79% of native TensorFlow speed with minimal overhead.
• Encapsulated TensorFlow API: Enjoy a cleaner and maintainable interface by abstracting TensorFlow's internals.

Performance Evaluation

Compared to the native TensorFlow C++ API, Txeo has been rigorously benchmarked using a multiclassification convolution model, demonstrating negligible performance overhead ranging from 0.65% to 1.21%.

Getting Started

Txeo facilitates quick setup with simple C++ examples. Below are two usage scenarios:

Example 1: Tensor Basics

This example showcases how to create, save, and load a tensor:

#include "txeo/Tensor.h"
#include "txeo/TensorIO.h"
#include <iostream>
using namespace txeo;
using namespace std;

int main() {
    Tensor<double> tensor({3, 3}, {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0});
    TensorIO::write_textfile(tensor, "tensor.txt");
    auto loaded_tensor = TensorIO::read_textfile<double>("tensor.txt");
    loaded_tensor.reshape({9});
    cout << loaded_tensor << endl;
    return 0;
}

Example 2: Running Inference with a Saved Model

This example demonstrates loading a model and running inference:

#include "txeo/Predictor.h"
#include "txeo/Tensor.h"
#include "txeo/TensorIO.h"

using namespace txeo;
#include <iostream>
#include <string>

int main() {
    string model_path = "path/to/saved_model";
    string input_path = "path/to/input_tensor.txt";
    string output_path = "path/to/output_tensor.txt";

    Predictor<float> predictor{model_path};
    auto input = TensorIO::read_textfile<float>(input_path);
    auto output = predictor.predict(input);
    TensorIO::write_textfile(output, output_path);
    return 0;
}

Future Development

Txeo is continuously evolving, with upcoming features focused on model training and advanced tensor operations, enabling capabilities like backpropagation support, gradient descent with various optimizers, and richer tensor manipulation techniques.

Equipped with user-friendly documentation and extensive examples, Txeo caters to both newcomers and seasoned developers looking to harness the power of TensorFlow through a modern C++ interface. For more information, visit the Txeo documentation at txeo site.

Txeo stands as a practical solution for developers aiming for efficiency and simplicity in C++ TensorFlow programming while maintaining robust performance.