Uploaded image for project: 'AI Platform Core Components'
  1. AI Platform Core Components
  2. AIPCC-8113

Skipping combo kernel for large pointwise nodes

    • Icon: Story Story
    • Resolution: Unresolved
    • Icon: Normal Normal
    • None
    • None
    • PyTorch
    • AIPCC Accelerators 21, AIPCC Accelerators 22, AIPCC Accelerators 23, AIPCC Accelerators 24, AIPCC Accelerators 25, AIPCC Accelerators 26

      For vllm + gpt-oss-20b + combo kernel, we can see ~50 prints of `ComboKernels: 1 large pointwise nodes are separated`. We should check if this misses some optimization opportunities, or consider not spam the user output.

      Repro and log: [P1991444071](https://www.internalfb.com/phabricator/paste/view/P1991444071)

      cc @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @aakhundov @coconutruben @zou3519 @eellison

              rh-ee-trichmon Turner Richmond
              rh-ee-trichmon Turner Richmond
              PyTorch Compile
              Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

                Created:
                Updated: