Nnabla-ext-cuda

release-note-bugfix

+ [fix: solve the nccl error while running more than 2 gpus](https://github.com/sony/nnabla-ext-cuda/pull/273)
+ [add code to implement asymmetric watch dog](https://github.com/sony/nnabla-ext-cuda/pull/271)
+ [update GNUmakefile in example/cpp](https://github.com/sony/nnabla-ext-cuda/pull/270)

release-note-build

+ [fix: solve the bug that no found CUDNN_PATH while building in windows](https://github.com/sony/nnabla-ext-cuda/pull/274)
+ [Fix failed to compile nnabla cuda wheel.](https://github.com/sony/nnabla-ext-cuda/pull/272)

release-note-core

+ [Support out-of-core training](https://github.com/sony/nnabla-ext-cuda/pull/268)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix

+ [CuDNN does not support fused batchnorm on compute_arch 5.3](https://github.com/sony/nnabla-ext-cuda/pull/262)

release-note-build

+ [Support cuDNN8](https://github.com/sony/nnabla-ext-cuda/pull/265)

release-note-op-layer

+ [BatchLogdet Function](https://github.com/sony/nnabla-ext-cuda/pull/266)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix
+ [Limit max threads to values per kernel function and device.](https://github.com/sony/nnabla-ext-cuda/pull/252)
+ [BugFix: Index of height and width in the channel-last of RandomErase function](https://github.com/sony/nnabla-ext-cuda/pull/250)

release-note-build
+ [Change dockerfile for runtime.](https://github.com/sony/nnabla-ext-cuda/pull/261)
+ [remove cudnn version from package name](https://github.com/sony/nnabla-ext-cuda/pull/263)
+ [Replace nnabla_tf_converter to nnabla_converter](https://github.com/sony/nnabla-ext-cuda/pull/264)
+ [Update dockerfile for CUDA10.2](https://github.com/sony/nnabla-ext-cuda/pull/260)
+ [update dockerfiles to install nnabla_tf_converter](https://github.com/sony/nnabla-ext-cuda/pull/259)
+ [Change libarchive location](https://github.com/sony/nnabla-ext-cuda/pull/258)
+ [Separate tf converter from nnabla.](https://github.com/sony/nnabla-ext-cuda/pull/257)
+ [Build with chocolatey environment.](https://github.com/sony/nnabla-ext-cuda/pull/251)

release-note-core
+ [Support cuDNN8](https://github.com/sony/nnabla-ext-cuda/pull/256)

release-note-op-layer
+ [Norm and NormNormalization](https://github.com/sony/nnabla-ext-cuda/pull/255)
+ [Add Cuda implementation of AdaBelief solver](https://github.com/sony/nnabla-ext-cuda/pull/254)
+ [Add cuda implementation of scatter_add function](https://github.com/sony/nnabla-ext-cuda/pull/253)
+ [Inplace for unary/binary arithmetic functions](https://github.com/sony/nnabla-ext-cuda/pull/249)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix
+ [Fix the compare function for reverse sorting.](https://github.com/sony/nnabla-ext-cuda/pull/246)
+ [Fix the CUDA transpose operation for large N-dimensional inputs.](https://github.com/sony/nnabla-ext-cuda/pull/245)

release-note-op-layer
+ [Add RMSpropGraves](https://github.com/sony/nnabla-ext-cuda/pull/248)
+ [Memory Optimization](https://github.com/sony/nnabla-ext-cuda/pull/247)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-utility
+ [extend watch dog time so not too sensitive](https://github.com/sony/nnabla-ext-cuda/pull/244)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix

+ [Fix the gradient operation when inplaced with output variable.](https://github.com/sony/nnabla-ext-cuda/pull/237)

release-note-build

+ [extend the timeout of all_reduce to avoid CI error](https://github.com/sony/nnabla-ext-cuda/pull/242)
+ [Upgrade gcc to 4.9.2](https://github.com/sony/nnabla-ext-cuda/pull/241)
+ [Update docker files](https://github.com/sony/nnabla-ext-cuda/pull/239)

release-note-op-layer

+ [Add Mish Activation](https://github.com/sony/nnabla-ext-cuda/pull/238)

release-note-utility

+ [add watch dog for those might cause dead-waiting functions](https://github.com/sony/nnabla-ext-cuda/pull/240)

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](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)