Version 1.9.0 Release

release-note-break-compat

• Change align_corners to False as default and add warning notes

release-note-bugfix

• Add **kw args to ref_dequantize_linear and ref_grad_dequantize_linear.
• Add FusedConvolution interface and some enhancements & fixes
• Fix log double backward

release-note-build

• Change default build environment from CentOS6 to CentOS7
• update devtoolset package version. update url of yum repository

release-note-doc

• Manual translation
• Correct the warp_by_flow documentation
• Add MNIST Tutorials
• Feature/20200529 fix documents
• Update nnabla convert doc at 20200521
• Add DCGAN Tutorial

release-note-format-converter

• Fix PReLU for ONNX Exporter.
• add support for convert nnabla to tflite
• Support deconv's output_padding for ONNX Exporter.
• Fix Broadcast for ONNX Exporter.

release-note-op-layer

• Quantize/DequantizeLinear
• Add deconvolution output padding option.
• Interpolation with half_pixel and half_pixel_for_nn

release-note-utility

• feat: save optimizer checkpoint in .h5 format
• Change cuDNN algorithm selection method to improve inference performance
• Add AdamW, Lars and SgdW to load.py
• feat: relocate cg_load info from log to progress.txt and status.json,…
• feat: save optimizer states every epoch
• Fix Unpooling for SNPE
• fix nnp to nnp problem with h5
• using sliced data set for monitor, instead of full data set for multi…
• Fix bug when reading h5 file in NNP
• Get device-id from local_rank
• remove thread-based watchdog for allreduce
• feat: measure cpu/gpu load & write to status.json when psutil, pynvml…

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 # For CUDA version 10.2 users

You can also install the cuda extension with specific versions from one of the following. See also FAQ

• nnabla-ext-cuda90 (CUDA 9.0 x cuDNN 7.6)
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.6)
• nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 7.6)

pip install nnabla
pip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 7.6 users

For distributed training, You need to install the correct package that
matches the version of MPI installed on your system.

We prepared following packages.

• nnabla-ext-cuda90-nccl2-mpi1-10-2 (Ubuntu16.04 default)
• nnabla-ext-cuda90-nccl2-mpi2-1-1 (Ubuntu18.04 default)
• nnabla-ext-cuda90-nccl2-mpi3-1-6
• nnabla-ext-cuda100-nccl2-mpi1-10-2 (Ubuntu16.04 default)
• nnabla-ext-cuda100-nccl2-mpi2-1-1 (Ubuntu18.04 default)
• nnabla-ext-cuda100-nccl2-mpi3-1-6
• nnabla-ext-cuda102-nccl2-mpi1-10-2 (Ubuntu16.04 default)
• nnabla-ext-cuda102-nccl2-mpi2-1-1 (Ubuntu18.04 default)
• nnabla-ext-cuda102-nccl2-mpi3-1-6

If you want to use a version of MPI not listed above, you need to build it from the source.

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

Build Guide
The "nnabla-ext-cuda" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda102, nnabla-ext-cuda100, nnabla-ext-cuda90 instead.
The following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.

• nnabla-ubuntu16
• nnabla-ubuntu18
• nnabla-ext-cuda91
• nnabla-ext-cuda92
• nnabla-ext-cuda101
• nnabla-ext-cuda90-nccl2-ubuntu16
• nnabla-ext-cuda100-nccl2-ubuntu16
• nnabla-ext-cuda100-nccl2-ubuntu18
• nnabla-ext-cuda101-nccl2-ubuntu16
• nnabla-ext-cuda101-nccl2-ubuntu18

The following "nnabla-ext-cuda" docker images have been deprecated.

• py27-cuda92
• py36-cuda92
• py37-cuda92
• py27-cuda92-v1.0.xx
• py36-cuda92-v1.0.xx
• py37-cuda92-v1.0.xx

We've decided to change nnabla's versioning policy to semantic versioning.
This change has been applied from version 1.1.0.

Python 2 is no longer be supported from v1.5.0.
CUDA8.0 is no longer be supported from v1.6.0.