Add tuned shapes for OBA 223x scaled model (#5053)

Doris Xiang · meta-codesync[bot] · commit 9b5af57c214e · 2025-10-24T14:42:54.000-07:00
Summary: Pull Request resolved: #5053 - full space tuning for OBA 223x scaled model Differential Revision: D85449482 fbshipit-source-id: b41858c0ffc52247134a45ccbf33ed2542b6de40
diff --git a/fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py b/fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py
@@ -3840,6 +3840,10 @@ def get_full_non_persistent_tuning_space():
     (256, 128, 128, 1, 1, 2, 16, 1, 8, 2),
     (128, 256, 128, 2, 1, 2, 16, 2, 4, 1),
     (256, 128, 64, 2, 1, 2, 16, 1, 4, 2),
+    (128, 128, 256, 2, 1, 0, 16, 2, 8, 2),
+    (128, 64, 128, 2, 1, 2, 16, 2, 4, 2),
+    (128, 128, 64, 2, 1, 0, 16, 1, 4, 2),
+    (128, 128, 128, 1, 1, 2, 16, 1, 4, 2),
 ]
 
 

Original file line number	Diff line number	Diff line change
`@@ -3840,6 +3840,10 @@ def get_full_non_persistent_tuning_space():`
`3840`	`3840`	`(256, 128, 128, 1, 1, 2, 16, 1, 8, 2),`
`3841`	`3841`	`(128, 256, 128, 2, 1, 2, 16, 2, 4, 1),`
`3842`	`3842`	`(256, 128, 64, 2, 1, 2, 16, 1, 4, 2),`
	`3843`	`+ (128, 128, 256, 2, 1, 0, 16, 2, 8, 2),`
	`3844`	`+ (128, 64, 128, 2, 1, 2, 16, 2, 4, 2),`
	`3845`	`+ (128, 128, 64, 2, 1, 0, 16, 1, 4, 2),`
	`3846`	`+ (128, 128, 128, 1, 1, 2, 16, 1, 4, 2),`
`3843`	`3847`	`]`
`3844`	`3848`
`3845`	`3849`