The RAPIDS Accelerator is distributed through the NGC Catalog and can be pulled directly via a browser. Manual upload of the Accelerator is required for Dataproc and Databricks deployments to ensure the correct version of the accelerator is used. EMR maintains a separate version. To access the RAPIDS Accelerator, navigate and log into NGC and access the NGC Catalog Select Feature Branches and Models.
Select Entities followed by the RAPIDS Accelerator for Apache Spark.
Under Version History Select the available version and the associated ellipsis to directly download the jar file.
The saved jar file will be uploaded to either a GCP bucket for Dataproc deployments or a Databricks File System (dbfs) folder for Databricks deployments.