[BugFix] Make max_batch_size work in Omni online serving (#487)

Signed-off-by: ZeldaHuang <hzm414167@alibaba-inc.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

tests/e2e/online_serving/stage_configs/qwen3_omni_ci.yaml

@@ -8,7 +8,7 @@ stage_args:
   - stage_id: 0
     runtime:
       devices: "0,1"
-      max_batch_size: 1
+      max_batch_size: 5
     engine_args:
       model_stage: thinker
       model_arch: Qwen3OmniMoeForConditionalGeneration
@@ -38,7 +38,7 @@ stage_args:
   - stage_id: 1
     runtime:
        devices: "1"
-       max_batch_size: 1
+       max_batch_size: 5
     engine_args:
        model_stage: talker
        model_arch: Qwen3OmniMoeForConditionalGeneration

vllm_omni/entrypoints/omni_stage.py

@@ -880,6 +880,8 @@ async def _stage_worker_async(
     except Exception as e:
         logger.warning("Device setup failed: %s", e)
 
+    max_batch_size = int(runtime_cfg.get("max_batch_size", 1) or 1)
+    engine_args["max_num_seqs"] = max_batch_size
     # Initialize OmniConnectors if configured to match sync worker behavior
     connectors: dict[Any, Any] = {}
     if connectors_config: