[AMD] CI - enable deepseekv3.2 on MI325-8gpu and merge perf/accuracy test suites into stage-b suites (#17633)

Co-authored-by: Bingxu Chen <Bingxu.Chen@amd.com>

test/registered/amd/test_deepseek_v32_basic.py

@@ -15,11 +15,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_amd_ci(
-    est_time=3600,
-    suite="stage-c-test-large-8-gpu-amd-mi35x",
-    disabled="move to nightly for saving time",
-)
+register_amd_ci(est_time=1800, suite="stage-c-test-large-8-gpu-amd")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 
@@ -159,7 +155,7 @@ class TestDeepseekV32TP(CustomTestCase):
                 f"### test_bs_1_speed (deepseek-v32)\n" f"{speed=:.2f} token/s\n"
             )
             if is_in_amd_ci():
-                self.assertGreater(speed, 20)
+                self.assertGreater(speed, 15)
             else:
                 self.assertGreater(speed, 70)
 

test/registered/amd/test_deepseek_v32_mtp.py

@@ -17,7 +17,12 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_amd_ci(est_time=3600, suite="stage-c-test-large-8-gpu-amd-mi35x")
+register_amd_ci(
+    est_time=3600,
+    suite="stage-c-test-large-8-gpu-amd-mi35x",
+    disabled="move to nightly for saving time",
+)
+
 FULL_DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 
 

test/registered/amd/test_kimi_k2_instruct.py

@@ -11,12 +11,13 @@ from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
+    is_in_amd_ci,
     is_in_ci,
     popen_launch_server,
     write_github_step_summary,
 )
 
-register_amd_ci(est_time=3600, suite="stage-c-test-large-8-gpu-amd-mi35x")
+register_amd_ci(est_time=3600, suite="stage-c-test-large-8-gpu-amd")
 
 KIMI_K2_MODEL_PATH = "moonshotai/Kimi-K2-Instruct-0905"
 SERVER_LAUNCH_TIMEOUT = 3600
@@ -88,7 +89,10 @@ class TestKimiK2Instruct0905(CustomTestCase):
                 f"### test_bs_1_speed (Kimi-K2-Instruct-0905)\n"
                 f"{speed=:.2f} token/s\n"
             )
-            self.assertGreater(speed, 45)
+            if is_in_amd_ci():
+                self.assertGreater(speed, 30)
+            else:
+                self.assertGreater(speed, 45)
 
 
 if __name__ == "__main__":