[MM] Skip VMM error gathers for text-only requests (#40005)

Co-authored-by: metamergebot <324680979+metamergebot@users.noreply.github.com>
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>

python/sglang/srt/managers/scheduler.py

@@ -2192,11 +2192,14 @@ class Scheduler(
 
         request_errors = []
         for tokenized_req in tokenized_reqs:
+            # The request broadcast makes this skip consistent across ranks.
+            if tokenized_req.mm_inputs is None:
+                request_errors.append(None)
+                continue
+
             local_error = None
             try:
-                if tokenized_req.mm_inputs is not None and not isinstance(
-                    tokenized_req.mm_inputs, MultimodalInputs
-                ):
+                if not isinstance(tokenized_req.mm_inputs, MultimodalInputs):
                     tokenized_req.mm_inputs = MultimodalInputs.from_processor_output(
                         tokenized_req.mm_inputs,
                         requires_mm_token_modalities=self.model_config.requires_mm_token_modalities,

test/registered/unit/multimodal/test_gpu_feature_transport.py

@@ -9,6 +9,7 @@ from unittest.mock import AsyncMock, MagicMock, call, patch
 import torch
 
 from sglang.test.ci.ci_register import register_cpu_ci
+from sglang.test.test_utils import CustomTestCase
 
 register_cpu_ci(est_time=11, suite="base-a-test-cpu")
 
@@ -760,7 +761,7 @@ class TestCudaVmmFeatureTransport(unittest.TestCase):
         self.assertIs(transport.pool, pool)
 
 
-class TestSchedulerMmTransportBoundary(unittest.TestCase):
+class TestSchedulerMmTransportBoundary(CustomTestCase):
     def _publish(self, **fields):
         from sglang.srt.runtime_context import get_context