mxnet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From sandeep krishnamurthy <...@apache.org>
Subject [Launch Announcement] Keras 2 with Apache MXNet (incubating) backend
Date Tue, 22 May 2018 16:15:06 GMT
Hello MXNet community,

Keras users can now use the high-performance MXNet deep learning engine for
the distributed training of convolutional neural networks (CNNs) and
recurrent neural networks (RNNs). With an update of a few lines of code,
Keras developers can increase training speed by using MXNet's multi-GPU
distributed training capabilities. Saving an MXNet model is another
valuable feature of the release. You can design in Keras, train with
Keras-MXNet, and run inference in production, at-scale with MXNet.

>From our initial benchmarks, CNNs with Keras-MXNet is up to 3X faster on
GPUs compared to the default backend. See the benchmark module
<https://github.com/awslabs/keras-apache-mxnet/tree/master/benchmark> for
more details.

RNN support in this release is experimental with few known
issues/unsupported functionalities. See using RNN with Keras-MXNet
limitations and workarounds doc
<https://github.com/awslabs/keras-apache-mxnet/blob/master/docs/mxnet_backend/using_rnn_with_mxnet_backend.md>
for more details.

See Release Notes
<https://github.com/awslabs/keras-apache-mxnet/releases/tag/v2.1.6> for
more details on unsupported operators and known issues. We will continue
our efforts in the future releases to close the gaps.

Thank you for all the contributors - Lai Wei <https://github.com/roywei>, Karan
Jariwala <https://github.com/karan6181/>, Jiajie Chen
<https://github.com/jiajiechen>, Kalyanee Chendke <https://github.com/kalyc>,
Junyuan Xie <https://github.com/piiswrong>

We welcome your contributions -
https://github.com/awslabs/keras-apache-mxnet. Here is the issue with the
list of operators to be implemented. Do check it out and create a PR -
https://github.com/awslabs/keras-apache-mxnet/issues/18

Best,
Sandeep

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message