fix: adjust aws_c5.18xlarge memory size (#991)

Co-authored-by: Alexander Zhipa <[email protected]>
pytorch · Mar 7, 2025 · 357b9d9 · 357b9d9
1 parent 767bc29
commit 357b9d9
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 2 deletions.
diff --git a/torchx/specs/named_resources_aws.py b/torchx/specs/named_resources_aws.py
@@ -142,7 +142,11 @@ def aws_m5_2xlarge() -> Resource:
 
 def aws_c5_18xlarge() -> Resource:
     return Resource(
-        cpu=72, gpu=0, memMB=144 * GiB, capabilities={K8S_ITYPE: "c5.18xlarge"}
+        # using lower memory size than the spec since MEM_TAX is not enough for adjustment
+        cpu=72,
+        gpu=0,
+        memMB=142 * GiB,
+        capabilities={K8S_ITYPE: "c5.18xlarge"},
     )
 
 

diff --git a/torchx/specs/test/named_resources_aws_test.py b/torchx/specs/test/named_resources_aws_test.py
@@ -249,7 +249,7 @@ def test_aws_c5_18xlarge(self) -> None:
         resource = aws_c5_18xlarge()
         self.assertEqual(72, resource.cpu)
         self.assertEqual(0, resource.gpu)
-        self.assertEqual(144 * GiB, resource.memMB)
+        self.assertEqual(142 * GiB, resource.memMB)
 
     def test_aws_t3_medium(self) -> None:
         resource = aws_t3_medium()