From cc1ca61c81fb2133ab134fd4294d11828fa89a9e Mon Sep 17 00:00:00 2001
From: Alison Shao <54658187+alisonshao@users.noreply.github.com>
Date: Wed, 25 Feb 2026 10:27:31 -0800
Subject: [PATCH] fix: add --cuda-graph-max-bs to DSV3 FA3 FP8 KV cache test
 (#19307)

---
 test/registered/mla/test_mla_deepseek_v3.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/test/registered/mla/test_mla_deepseek_v3.py b/test/registered/mla/test_mla_deepseek_v3.py
index b7518ede6..c30d6827b 100644
--- a/test/registered/mla/test_mla_deepseek_v3.py
+++ b/test/registered/mla/test_mla_deepseek_v3.py
@@ -110,7 +110,9 @@ class TestMLADeepseekV3Fa3Fp8Kvcache(CustomTestCase):
             "fp8_e4m3",
         ]
         if is_cuda():
-            other_args.extend(["--attention-backend", "fa3"])
+            other_args.extend(
+                ["--attention-backend", "fa3", "--cuda-graph-max-bs", "2"]
+            )
         cls.process = popen_launch_server(
             cls.model,
             cls.base_url,