Brezel is a high-performance tensor computation library that combines the elegance of PyTorch's API with the performance benefits of modern C++20. Designed for researchers, engineers, and data scientists who need efficient tensor operations without sacrificing usability.

auto a = bz::tensor::ones({2, 3});
auto b = bz::tensor::arange(6).reshape({2, 3});

// Perform operations
auto c = a * b + brezel::sin(b);

// Neural Network Operations
auto model = bz::nn::Sequential(
  bz::nn::Linear(784, 128),
  bz::nn::ReLU(),
  bz::nn::Linear(128, 10)
)

auto predictions = model->forward(input);

• 🚀 Performance-First: Optimized core using SIMD, multi-threading, and GPU acceleration
• 🔧 Modern C++20: Leveraging concepts, ranges, and other modern C++ features
• 🧠 Deep Learning Primitives: Neural network building blocks with automatic differentiation
• 📊 Data Science Tools: Statistical functions, data loading, and visualization utilities
• 📦 No Dependencies: Core functionality has minimal external dependencies
• 🔌 Extensibility: Easy to add custom operations and backends