Enable Priority Configs (All Gather) (Master) #11734
Assignees
Labels
feature
op_cat: ccl
Op Generalization
Generalization and relaxations of requirements in Ops
P1
perf
for issues tracking performance problems/improvements
Comments
This was referenced Aug 21, 2024
SeanNijjar
added
P1
feature
Op Generalization
|
These test cases are running in TG post-commit (frequent) already |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Tasks
Priority Configs:
Prior to sweeping, the priority configs should be added and tested.
From Llama 405B
4chips,3links, input tensor (per chip) = [1,1,32,6.5*1024], dim=1 => input_tensor for test_case =[1,4,32,6.5*1024])8chips,{3,4}links, input tensor (per chip) = [1,1,32,2304], dim=1 => input_tensor for test_case =[1,8,32,2304])8chips,{3,4}links, input tensor (per chip) = [1,1,32,4k], dim=1 => input_tensor for test_case =[1,8,32,4k])8chips,{3,4}links, input tensor (per chip) = [1, 1, 8[padded to 32], 4k], dim=2 => output shape (per chip) = [1, 1, 32, 4k] -> all-gather concatenates within tile8chips,{3,4}links, input tensor (per chip) = [1, 1, 8[padded to 32], 16k], dim=2 => output shape (per chip) = [1, 1, 32, 16k] -> all-gather concatenates within tileFrom Llama 70B
4chips,3links, input tensor (per chip) = [1,1,32,3.5*1024], dim=1 => input_tensor for test_case =[1,4,32,(int)(3.5*1024)])8chips,{3,4}links, input tensor (per chip) = [1,1,32,1280], dim=1 => input_tensor for test_case =[1,8,32,1280])8chips,{3,4}links, input tensor (per chip) = [1,1,32,2048], dim=1 => input_tensor for test_case =[1,8,32,2048])8chips,{3,4}links, input tensor (per chip) = [1, 1, 8[padded to 32], 2k], dim=2 => output shape (per chip) = [1, 1, 32, 2k] -> all-gather concatenates within tile8chips,{3,4}links, input tensor (per chip) = [1, 1, 8[padded to 32], 4k], dim=2 => output shape (per chip) = [1, 1, 32, 4k] -> all-gather concatenates within tileThe text was updated successfully, but these errors were encountered: