Based on feedback from our Caffe deep learning framework customers we've added pycaffe interface support to our AWS Caffe AMIs. Pycaffe is the python interface for Caffe. When you import caffe into your python scripts you can use it to load models, do forward and backward propagation, handle IO, visualize networks, and even perform model solving. All model data, derivatives, and parameters are exposed through the interface for reading and writing.
Currently we have two Caffe AMIs with pycaffe support in the AWS Marketplace:
The first AMI, Bitfusion Ubuntu 14 Cuda 7 Caffe, is pre-configured as a standalone AMI and is optimized for GPU instances available in AWS. It comes pre-installed wit Nvidia Drivers, the Cuda Toolkit, the CuDNN R3 library, and demo collateral.
The second one, Bitfusion Boost Client Caffe, is pre-configured to work flawlessly with our Bitfusion Boost Server AMI, allowing you to run multiple Caffe clients on separate instances which share a single GPU instance for maximum utilization, or to run a single client which utilizes multiple GPU servers for maximum performance. Detailed instructions on setting up the Server/Client configuration can be found here.
What else can we do to improve our AMIs and make you more productive? Drop us a note and we'll be glad to consider it for our next release.