Version 1.0.19 Release

release-note-bugfix

• Fix cuda kernel block calculation for zero size.
• [fix] Set mathType properly in CUDNN Convolution

release-note-build

• Fix test on ppc64le
• Add ppc64le build support.

release-note-op-layer

• Add 1-D linear and nearest neighbor interpolation.
• Add CUDA implementation of scatter_nd function

release-note-utility

• [Solver] AdamW and SGDW
• LARS (Layer-wise Adaptive Rate Scaling)

Install the latest nnabla by:

pip install nnabla
pip install nnabla_ext_cuda  # For CUDA users

Users with python <= 3.4 may experience errors with pip install nnabla and pip install nnabla-ext-cuda.
■ Workaround
Please install matplotlib == 2.2.3 and re-install nnabla, nnabla_ext_cuda.

pip install matplotlib==2.2.3
pip install nnabla
pip install nnabla_ext_cuda

Note that CUDA 10.1 and cuDNN 7.6 are set as default if versions are not specified. You can also install the cuda extension with specific versions from one of the following. See also FAQ

• nnabla-ext-cuda80 (CUDA 8.0 x cuDNN 7.1)
• nnabla-ext-cuda90 (CUDA 9.0 x cuDNN 7.6)
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.6)
• nnabla-ext-cuda101 (CUDA 10.1 x cuDNN 7.6)

pip install nnabla
pip install nnabla_ext_cuda101  # For CUDA 10. x cuDNN 7.6 users

Additional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.

Install Guide

To use C++ inference feature, follow the demonstration on MNIST inference in C++.

Demonstration

For distributed training, you need to build a binary from source. See the guide for building multi-GPU training package.

Build Guide