I’m trying to test the new Elastic EP feature and have a few questions regarding the correct setup and testing procedure.
In the examples/online_serving/elastic_ep/ directory, I found several scripts. Are these intended to be used directly for testing Elastic EP functionality? If so, is there any specific order or configuration required to run them successfully?
I also noticed the script tools/ep_kernels/elastic_ep/install_eep_libraries.sh. Is it necessary to run this script before launching any Elastic EP workloads? Additionally, this script appears very similar to tools/ep_kernels/install_python_libraries.sh, which installs DeepEP dependencies. Could you clarify the relationship between these two scripts? Specifically:
Does running install_eep_libraries.sh supersede the need to run install_python_libraries.sh?
Or should both scripts be executed, and if so, in what order?
Any guidance on the proper testing workflow and dependency setup for Elastic EP would be greatly appreciated. Thank you!
It seems likely that the scripts in examples/online_serving/elastic_ep/ are intended as reference examples for testing Elastic EP, and you can use them directly to validate functionality. There is no explicit documentation in the data sources specifying a required order, but generally, you should ensure all dependencies are installed before running these examples. As for the installation scripts, tools/ep_kernels/install_python_libraries.sh installs the core EP dependencies (DeepEP, pplx-kernels), while tools/ep_kernels/elastic_ep/install_eep_libraries.sh appears to be a specialized script for Elastic EP, possibly including additional or updated components for elastic scaling. However, the data sources do not explicitly state whether running install_eep_libraries.sh supersedes install_python_libraries.sh, or if both are required.
My understanding is that you should run both scripts to ensure all dependencies are met, starting with install_python_libraries.sh (for base EP support), followed by install_eep_libraries.sh (for Elastic EP-specific components), unless the Elastic EP script’s documentation says otherwise. There is no clear statement in the data sources about one script replacing the other, so running both in order is the safest approach. Would you like a step-by-step breakdown of the setup and testing workflow based on the available examples and scripts?