Recipes for creating Tensorflow conda packages.
In the defaults channel Tensorflow is provided via a number of packages.
As of version 1.10.0, the library itself is provided by the tensorflow-base
package. On Windows and Mac the tensorflow-base recipe is used to produce
this package. On Linux, three different variants of the tensorflow-base
package are provided, a variant which uses MKL, a variant which uses Eigen, and
a GPU variant. These are created using the tensorflow-base-cpu,
and tensorflow-base-gpu recipes.
Installing the tensorflow package using conda installs both the tensorflow
library as well as tensorboard. The tensorboard recipe is used to create
the tensorboard package.
The tensorflow metapackage package is created by the tensorflow recipe.
On Windows and Mac a single tensorflow metapackage is used which depends on
the correct versions of the tensorflow-base and tensorboard packages.
On Linux, the tensorflow metapackage depends on tensorboard, an exact
build of tensorflow-base and the version of the _tflow_1100_select package
which matches the tensorflow-base variant. The _tflow_1100_select package,
created from the _tflow_1100_select recipe, establishes the priority of the
variants using the version number. The variant with the highest version will
be installed by default. The non-default variant can be installed using the
tensorflow-mkl, tensorflow-eigen and tensorflow-gpu packages which are
created from the tensorflow-variants recipe.
Available Recipe:
- tensorboard : Tensorboard.
- tensorflow : Metapackage which installs tensorflow-base and tensorboard.
- tensorflow-base : The Tensorflow library. Used on Windows and Mac.
- tensorflow-base-cpu : Eigen and MKL variants of the Tensorflow library, Linux only.
- tensorflow-base-gpu : GPU variant of the Tensorflow library, Linux only.
- tensorflow-variants : Recipe used to create tensorflow variant packages, e.g. tensorflow-mkl.
- _tflow_1100_select : Metapackage to establish priority in tensorflow-base packages.