You can run it 'manually' as a YARN app in CDH 5.4. It should largely Just Work.
As you can make out here, CDH 5.5 is actually already based on what will be Spark 1.5:
CDH 5.4 can't update a minor release to 1.4 in a maintenance release, but 5.5 looks like it will jump to 1.5.
We do not have any instructions available since we do not support it. Spark has a Hadoop, among others, dependency and it needs to be compiled against the correct version of Hadoop.
You would need to create your own Spark install: compile Spark against the CDH 5.4 artifacts and make sure you reference the correct dependencies.