Caffe framework

Z MetaCentrum
Přejít na: navigace, hledání

Description

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by the Berkeley Vision and Learning Center (BVLC) and by community contributors.

Category

Development tools and environments

Availability

Version 2016 (September)

Licence

Caffe is released under the BSD 2-Clause license.

Use

MetaCentrum provides several version of caffe framework:

CPU version: module add caffe2016-cpu
GPU version: module add caffe2016-gpu
GPU + cuDNN version: module add caffe2016-gpu-cudnn

Please note that the GPU version is faster than the CPU version.

Moreover, testing of our MetaCentrum builds has shown that the GPU version compiled with cuDNN library is several times faster than the GPU version without it.

However, cuDNN is only supported on doom and zubat clusters (see GPU_clusters#FAQ), on other GPU clusters it is only possible to run the GPU version without it.

Our versions of Caffe include pycaffe interface. After loading Caffe module type

python
import caffe

and you can immediately start.

Documentation

http://caffe.berkeleyvision.org/tutorial/

Homepage

http://caffe.berkeleyvision.org/

Program manager

meta@cesnet.cz