DO NOT DIRECTLY EDIT THIS TICKET.

WARNING: THIS TICKET IS PUBLIC.  THE RESULTING CLONED TICKET'S SECURITY LEVEL CAN BE CHANGED IF NECESSARY.

This template ticket can be cloned by developers and contributors to Red Hat's AI projects to request that new or updated packages be included in the AIPCC builder system.  Please follow the instructions below.

Instructions:

1. Read the attached [^Red Hat AI Python Packaging Best Practices.pdf] to verify that your package meets the AIPCC team's packaging requirements. Packages not meeting these requirements will be rejected.
2. Run "AIPCC Clone Epics and Children" (under the 'More' menu) to duplicate this template ticket and its children. (Most Jira users do not have permission to do that using the default  "Clone Epic and Children" operation.)
3. Update the "Reporter" field to your name instead of the owner of the cloned ticket.
4. Update the <name> field in the Summary of the cloned ticket.
5. Update the <name> field in the Epic Name of the cloned ticket.
6. Fill out the “Requested Package Name and Version”, “Brief Explanation for request”, "QE user acceptance tests", and "Package License" in the cloned ticket.
7. If available, please include a link to the upstream package.  If necessary, the assignee of the cloned ticket may want to discuss alternative solutions in the upstream ticket and/or leave a comment about the resolution of the cloned ticket. 

Please use one package per ticket.

The ticket assignee may request additional information in the cloned ticket.

The text above this line can be deleted from the cloned ticket.

Requested Package Name and Version:

Give the exact requirement specification of the package as it appears in your requirements.txt or pyproject.toml, including the name, any extras specifier and the version. If you know the GitHub or gitlab repository that hosts the source, including that link will be helpful, too. For example:

nvshmem>=3.4.5

Brief Explanation for request

Upstream llm-d core images are moving to nvshmem 3.4.5, which are the initial github releases: https://github.com/NVIDIA/nvshmem

The main rationale is to not have to carry patches for IB device name overrides.

QE user acceptance tests

We need to test e2e, as testing requires things like IBGDA initialization. It's easiest to test on a live cluster with known good networking confguration.

Package License

This license has already been vetted by legal, but posting here: https://github.com/NVIDIA/nvshmem/blob/devel/License.txt