fix(ci): apply MMMU retry logic to all affected test files (#17329)

test/manual/nightly/test_vlms_piecewise_cuda_graph.py

@@ -3,12 +3,12 @@ import glob
 import json
 import os
 import random
-import subprocess
 import sys
 import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.kits.mmmu_vlm_kit import _run_lmms_eval_with_retry
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
@@ -85,11 +85,7 @@ class TestVLMPiecewiseCudaGraph(CustomTestCase):
             str(output_path),
         ]
 
-        subprocess.run(
-            cmd,
-            check=True,
-            timeout=3600,
-        )
+        _run_lmms_eval_with_retry(cmd, timeout=3600)
 
     def _run_vlm_mmmu_test(
         self,

test/manual/nightly/test_vlms_vit_cuda_graph.py

@@ -3,12 +3,12 @@ import glob
 import json
 import os
 import random
-import subprocess
 import sys
 import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.kits.mmmu_vlm_kit import _run_lmms_eval_with_retry
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
@@ -88,11 +88,7 @@ class TestVLMViTCudaGraph(CustomTestCase):
             str(output_path),
         ]
 
-        subprocess.run(
-            cmd,
-            check=True,
-            timeout=3600,
-        )
+        _run_lmms_eval_with_retry(cmd, timeout=3600)
 
     def _run_vlm_mmmu_test(
         self,

test/registered/vlm/test_encoder_dp.py

@@ -3,13 +3,13 @@ import glob
 import json
 import os
 import random
-import subprocess
 import sys
 import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.kits.mmmu_vlm_kit import _run_lmms_eval_with_retry
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
@@ -94,11 +94,7 @@ class TestVLMEncoderDP(CustomTestCase):
             str(output_path),
         ]
 
-        subprocess.run(
-            cmd,
-            check=True,
-            timeout=3600,
-        )
+        _run_lmms_eval_with_retry(cmd, timeout=3600)
 
     def _run_vlm_mmmu_test(
         self,

test/srt/test_epd_disaggregation.py

@@ -1,9 +1,9 @@
 import os
-import subprocess
 import threading
 import unittest
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.kits.mmmu_vlm_kit import _run_lmms_eval_with_retry
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
 )
@@ -177,7 +177,7 @@ class TestEPDDisaggregationOneEncoder(PDDisaggregationServerBase):
             limit,
         ]
 
-        subprocess.run(cmd, check=True, timeout=3600)
+        _run_lmms_eval_with_retry(cmd, timeout=3600)
 
     def test_mmmu(self):
         """Test MMMU evaluation with EPD disaggregation"""
@@ -393,7 +393,7 @@ class TestEPDDisaggregationMultiEncoders(PDDisaggregationServerBase):
             limit,
         ]
 
-        subprocess.run(cmd, check=True, timeout=3600)
+        _run_lmms_eval_with_retry(cmd, timeout=3600)
 
     def test_mmmu(self):
         """Test MMMU evaluation with EPD disaggregation (multiple encoders)"""