[
https://issues.apache.org/jira/browse/SINGA-359?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16473116#comment-16473116
]
ASF subversion and git services commented on SINGA-359:
-------------------------------------------------------
Commit 00573e26b6aafc1f3c8852bc3397c3a007cd2c89 in incubator-singa's branch
refs/heads/master from Wang Wei
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=00573e2 ]
SINGA-359 Create alias for conda packages
Move conda build scripts into tool/conda/singa folder as conda-build
expect to work in a flat folder.
> Create alias for conda packages
> -------------------------------
>
> Key: SINGA-359
> URL: https://issues.apache.org/jira/browse/SINGA-359
> Project: Singa
> Issue Type: Improvement
> Reporter: wangwei
> Priority: Major
>
> The conda packages stored on [anaconda
> cloud|https://anaconda.org/nusdbsystem/singa/files] are differentiated by the
> build string, which consists of the python version, cuda and cudnn version.
> The full name is like
> {code:java}
> singa=1.1.1=py36_cuda9.0_cudnn7.1.2{code}
> This ticket creates dummy conda packages with simple names like singa-cpu,
> singa-gpu. By enforcing the dependency on the real packages. e.g.
> {code:java}
> singa-gpu --> singa=1.1.1=py36_cuda9.0_cudnn7.1.2
> singa-cpu --> singa=1.1.1=py36_cpu
> singa --> singa-cpu{code}
> when we run
> {code:java}
> conda install -c nusdbsystem singa{code}
> the following package will be installed automatically
> {code:java}
> singa=1.1.1=py36_cpu{code}
> The new dummy packages are created in the similar way as (no need to have the
> build.sh)
> [https://github.com/apache/incubator-singa/tree/master/tool/conda]
> The dependency is specified in the run section, e.g.
> {code:java}
> run:
> - singa=1.1.1=py36_cpu{code}
> The channel needs to be added before running the build instruction.
> {code:java}
> conda config --add channels nusdbsystem{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
