primitiv 0.3.1.dev128

primitiv: A Neural Network Toolkit. (Python frontend)

Features

• Dynamic and incremental graph construction

• On-demand memory allocation

• Automatic minibatch broadcasting

• Mostly device-independent

• Simple usage

Install

Prerequisites:

• Python 3 (3.5 or later)

• NumPy (1.11.0 or later)

• Cython (0.27 or later)

• scikit-build (0.6.1 or later, only for building)

• (optional) CUDA (7.5 or later)

• (optional) OpenCL (1.2 or later) and OpenCL C++ binding v2

Install dependencies:

pip3 install numpy cython scikit-build

Build and install primitiv without CUDA and OpenCL:

pip3 install primitiv

Build and install primitiv with CUDA and/or OpenCL support:

# Enable only CUDA
pip3 install primitiv --global-option --enable-cuda

# Enable both CUDA and OpenCL
pip3 install primitiv --global-option --enable-cuda --global-option --enable-opencl

Notes

According to the manylinux1 policy described in PEP 513, binary packages are required to depend only on an extremely limited set of external shared libraries. Most users may install primitiv with CUDA and/or OpenCL backends that are not supported in the manylinux1 policy. For now, we provide only a source pacakge.

Resources

• Homepage

• Examples