Skip to content

add RoPE kernel to roofline plots in the latency-demo notebook #167

Description

@yuhaohaoyu

Lineage and Dependency

what's the goal

Add synthetic latency experiments of RoPE multiple core kernel in to the roofline sections of the latency_demo notebook.

What's the benefit

RoPE is an important accuracy boosting kernel used in IBM's Granite, and nearly every since DeepSeek. Including it in ktir-cpu's latency demo notebook is a great addition to showcase the usefulness of ktir-cpu.

How to assess completion

  1. Update codes under notebooks/, with codes to parameterization of the kernel to full fill the 4-core, 32-core-strong-scaling, and 32-core-weak-scaling setup for the roofline section of latency_demo notebook.
  2. Notebook run output (capture into html) for visual verification.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions