Aakaar
A high-performance, custom-built deep learning tensor library with a dynamic autograd engine and native C++/CUDA hardware acceleration.
Built from the ground up, Aakaar bridges the gap between Python's ease of use and C++'s execution speed, providing a PyTorch-like API for tensor manipulation, automatic differentiation, and neural network construction.
Key Features¶
- Dynamic Autograd Engine: Automatically constructs computational graphs on the fly and computes exact gradients via reverse-mode automatic differentiation.
- Hardware Acceleration: Seamlessly dispatches mathematical operations to optimized OpenBLAS (CPU) or cuBLAS (GPU) backends.
- Standalone CUDA Support: GPU acceleration works out-of-the-box. Pre-compiled wheels bundle the necessary CUDA runtime libraries, meaning no CUDA Toolkit installation is required for end-users.
- Familiar API: Designed to be intuitive for users familiar with modern deep learning frameworks, featuring
aa.Tensor,aa.nn.Linear, andaa.optim.SGD.
Installation¶
Aakaar provides pre-compiled wheels for Python 3.10 through 3.14 on both Windows and Linux.