Why does vLLM use eager mode based execution for CPUs #10716
                  
                    
                      amd-lalithnc
                    
                  
                
                  announced in
                General
              
            Replies: 0 comments
  
    Sign up for free
    to join this conversation on GitHub.
    Already have an account?
    Sign in to comment
  
        
    
Uh oh!
There was an error while loading. Please reload this page.
-
Looking through the code, I figured vLLM uses eager mode based execution for models.
I understand most ops are rewritten and replaced in the source model, are there any advantages or disadvantages to using eager mode over inductor path/torchscript path?
Beta Was this translation helpful? Give feedback.
All reactions