Closed pcmoritz closed 8 years ago
We provide JARs for CentOS 6 which make it possible to use Caffe with CPUs.
To use them, replace the relevant lines in your build.sbt
with the following ones:
resolvers += "javacpp" at "http://www.eecs.berkeley.edu/~rkn/snapshot-2016-03-05/"
resolvers += "javacpp" at "http://www.eecs.berkeley.edu/~rkn/snapshot-2016-03-23-CENTOS6-CPU/"
libraryDependencies += "org.bytedeco" % "javacpp" % "1.2-SPARKNETCENTOS6"
libraryDependencies += "org.bytedeco.javacpp-presets" % "caffe" % "rc3-1.2-SPARKNETCENTOS6"
libraryDependencies += "org.bytedeco.javacpp-presets" % "caffe" % "rc3-1.2-SPARKNETCENTOS6" classifier "linux-x86_64"
libraryDependencies += "org.bytedeco.javacpp-presets" % "opencv" % "3.1.0-1.2-SPARKNETCENTOS6"
libraryDependencies += "org.bytedeco.javacpp-presets" % "opencv" % "3.1.0-1.2-SPARKNETCENTOS6" classifier "linux-x86_64"
libraryDependencies += "org.bytedeco.javacpp-presets" % "tensorflow" % "master-1.2-SPARKNET"
libraryDependencies += "org.bytedeco.javacpp-presets" % "tensorflow" % "master-1.2-SPARKNET" classifier "linux-x86_64"
and uncomment
test in assembly := {}
to make sure tests are not run during assembly (the TensorFlow tests would not work on CentOS 6).
There are instructions in https://github.com/amplab/SparkNet/commit/61b0cbc8c2d5b5b26f64bf130c4cea6c49a24808 merged in #113 on how these JARs were build, in case you need to create your own.
see https://groups.google.com/forum/#!topic/sparknet-users/NBpa6EvEQW4