Feature Overview (mandatory - Complete while in New status)

Liger Kernel is a collection of Triton kernels designed specifically for LLM training. It can effectively increase multi-GPU training throughput by 20% and reduce memory usage by 60%.

IBM Cloud is looking for improvements to training time overhead in their SaaS. Liger kernels (https://github.com/linkedin/Liger-Kernel) have been shown to improve performance of training workloads by 20-30% as a result of significant memory reduction and GPU throughput improvement. 

Goals
RHEL AI users see improved training times out of the box as a result of Liger kernels

Requirements:

• Create an option to enable / disable the use of Liger kernels
• Provided experimental tests show improvement and there are no side effects, set the default of that option to true

Done - Acceptance Criteria:

An option for enabling Liger kernels is present in the ilab config file 

Use Cases - i.e. User Experience & Workflow:

(Initial completion while in Refinement status):
n/a

Out of Scope _{}(Initial completion while in Refinement status):{_}
n/a

Documentation Considerations _{}(Initial completion while in Refinement status):{_}
Provide information that needs to be considered and planned so that documentation will meet customer needs. If the feature extends existing functionality, provide a link to its current documentation..

In the ideal scenario, this is a setting that a user would not need to worry about. However reference documentation listing it and what it does would be appropriate

Questions to Answer _{}(Initial completion while in Refinement status):{_}
What gains can we expect to see with this optimization?

Background and Strategic Fit (Initial completion while in Refinement status):
Improving performance and bringing down training times improves the experience for our customers.

Customer Considerations _{}(Initial completion while in Refinement status):{_}
n/a