[CI] tune the test order to warmup the server (#5860)

python/sglang/srt/layers/quantization/deep_gemm.py

@@ -293,7 +293,7 @@ def _maybe_compile_deep_gemm_one_type_all(
         logger.info(
             f"Try DeepGEMM JIT Compiling for "
             f"<{kernel_helper.name}> N={n}, K={k}, num_groups={num_groups} with all Ms."
-            f"{' It only takes a litte time(Typically 1 sec) if you have run `sglang.compile_deep_gemm`. ' if not _IN_PRECOMPILE_STAGE else ''}"
+            f"{' It only takes a litte time (typically 1 sec) if you have run `python3 -m sglang.compile_deep_gemm`. ' if not _IN_PRECOMPILE_STAGE else ''}"
         )
 
         # NOTE(alcanderian): get_num_sms should be change when 2-batch-overlap is introduced

test/srt/test_full_deepseek_v3.py

@@ -35,7 +35,9 @@ class TestDeepseekV3(CustomTestCase):
     def tearDownClass(cls):
         kill_process_tree(cls.process.pid)
 
-    def test_gsm8k(self):
+    def test_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
         args = SimpleNamespace(
             num_shots=8,
             data_path=None,
@@ -100,7 +102,9 @@ class TestDeepseekV3MTP(CustomTestCase):
     def tearDownClass(cls):
         kill_process_tree(cls.process.pid)
 
-    def test_gsm8k(self):
+    def test_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
         requests.get(self.base_url + "/flush_cache")
 
         args = SimpleNamespace(