Version 1.2.0 Release

release-note-bugfix

• fix convolution segv caused by cudaEvent_t instance variables
• Generate random seed when seed==-1 in cudnn context

release-note-build

• Includes cuda lib in wheel

release-note-op-layer

• Add CUDA implementation of OneHot function
• RandomFlip implementation in CUDA

release-note-utility

• Skip allreduce if array is not updated

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda101  # For CUDA version 10.1 users

Users with python <= 3.4 may experience errors with pip install nnabla and pip install nnabla-ext-cuda101.
■ 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-cuda101

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 10.1 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

The "nnabla-ext-cuda" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda101, nnabla-ext-cuda100, nnabla-ext-cuda90 or nnabla-ext-cuda80 instead.
The following nnabla CUDA extension packages have been deprecated and the PyPi repository has been closed.

• nnabla-ext-cuda91
• nnabla-ext-cuda92
  The following "nnabla-ext-cuda" docker images have been deprecated.
• py37-cuda92
• py36-cuda92
• py27-cuda92
• py37-cuda92-v1.0.xx
• py36-cuda92-v1.0.xx
• py27-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.

Version 1.1.0 Release

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda101  # For CUDA version 10.1 users

Users with python <= 3.4 may experience errors with pip install nnabla and pip install nnabla-ext-cuda101.
■ 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-cuda101

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

The "nnabla-ext-cuda" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda101, nnabla-ext-cuda100, nnabla-ext-cuda90 or nnabla-ext-cuda80 instead.
The following nnabla CUDA extension packages have been deprecated and the PyPi repository has been closed.

• nnabla-ext-cuda91
• nnabla-ext-cuda92
  The following "nnabla-ext-cuda" docker images have been deprecated.
• py37-cuda92
• py36-cuda92
• py27-cuda92
• py37-cuda92-v1.0.xx
• py36-cuda92-v1.0.xx
• py27-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.

Version 1.0.20 Release

release-note-op-layer

• Concurrent streams in backward conv & disable nan propagation in MaxPoolingCudnn

release-note-utility

• Add CUDA stream handler class in Python (as an unsupported feature so far)
• Add NVTX utilities in Python

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

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

Version 1.0.18 Release

release-note-break-compat

• [Fix] LSTM/GRU Output Shape

release-note-bugfix

• [Fix] Derivatives for Activation Layers
• [Fix] PF.lstm, PF.gru weight param mismatch

release-note-build

• Fix compile errors in nbla_cli build

release-note-op-layer

• Add 2-D and 3-D nearest neighbor interpolation.
• [Fix] log_sigmoid computation efficiency
• Add CUDNN LogSoftmax and disable fp32 Softmax
• Add CUDNN Fused Batch Normalization and utilize faster CUDNN Batch Normalization
• CUDNN Convolution and Pooling w/ NHWC layout
• Add 3D linear interpolation support.

release-note-utility

• Fix AdaBound & AMSBound for learning rate scheduling (sony/nnabla#443)

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 9.2 and cuDNN 7.4 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 7.5(Linux), 7.6(Win))
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.5(Linux), 7.6(Win))
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.5(Linux), 7.6(Win))
• nnabla-ext-cuda101 (CUDA 10.1 x cuDNN 7.5(Linux), 7.6(Win))

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.4 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

Version 1.0.17 Release

• Support ubuntu18.04 for multi-gpu build.
• [Fix] RNN/LSTM/GRU Memory Leakage
• gather_nd and selection by index variable (advanced indexing)
• Fix bugs in min and max functions when used to return indices.
• Synchronized Batch Normalization
• Various Activation Funcs (hard_sigmoid, hard_tanh, log_sigmoid, relu6, softplus, softsign, tanh_shrink, sinc)

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 9.2 and cuDNN 7.4 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 7.5)
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.5)
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.5)

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.4 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

Version 1.0.16 Release

• Add numeric include
• Fix numpy requirements
• Add AdaBound & AMSBound
• Add tile function (c.f. numpy.tile or torch.repeat)
• Add CUDA implemention for random_choice function.
• Add TopKDataCuda and TopKGradCuda function implementations.

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 9.2 and cuDNN 7.4 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 7.5(win), 7.4(linux))
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.5(win), 7.4(linux))
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.5)

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.4 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

Version 1.0.15 Release

• [Bug] Image Augmentation

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 9.2 and cuDNN 7.3 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 7.3(win), 7.4(linux))
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.3(win), 7.4(linux))
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.3(win), 7.4(linux))

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.3 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

Version 1.0.14 Release

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 9.2 and cuDNN 7.3 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 7.3(win), 7.4(linux))
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.3(win), 7.4(linux))
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.3(win), 7.4(linux))

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.3 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

Version 1.0.13 Release

• Improve memory allocator
• [Function] Add GELU Activation
• [Fix] Minor English errors
• Change Docker image names
• C++ computation graph building APIs
• Apply source code format rules.

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 9.2 and cuDNN 7.3 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 7.3(win), 7.4(linux))
• nnabla-ext-cuda92 (CUDA 9.2 x cuDNN 7.3(win), 7.4(linux))
• nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.3(win), 7.4(linux))

pip install nnabla
pip install nnabla_ext_cuda92  # For CUDA 9.2 x cuDNN 7.3 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