fix connector metadata

vllm/v1/core/sched/output.py

@@ -155,3 +155,4 @@ class SchedulerOutput:
 
     # KV Cache Connector metadata.
     kv_connector_metadata: Optional[KVConnectorMetadata] = None
+    kv_lmcache_connector_metadata: Optional[KVConnectorMetadata] = None

vllm/v1/core/sched/scheduler.py

@@ -89,6 +89,7 @@ class Scheduler(SchedulerInterface):
             self.connector = KVConnectorFactory.create_connector_v1(
                 config=self.vllm_config, role=KVConnectorRole.SCHEDULER)
 
+        self.lmcache_connector = None
         if envs.VLLM_LMCACHE_ENABLE:
             lmcache_config = copy.deepcopy(self.vllm_config)
             lmcache_config.kv_transfer_config = KVTransferConfig(kv_connector="LMCacheConnectorV1", kv_role="kv_both")
@@ -602,7 +603,7 @@ class Scheduler(SchedulerInterface):
 
         if self.lmcache_connector is not None:
             meta = self.lmcache_connector.build_connector_meta(scheduler_output)
-            scheduler_output.kv_connector_metadata = meta
+            scheduler_output.kv_lmcache_connector_metadata = meta
 
         events = self.kv_cache_manager.take_events()
         if events:
@@ -1034,7 +1035,7 @@ class Scheduler(SchedulerInterface):
 
         if self.lmcache_connector is not None:
             meta = self.lmcache_connector.build_connector_meta(scheduler_output)
-            scheduler_output.kv_connector_metadata = meta
+            scheduler_output.kv_lmcache_connector_metadata = meta
 
         events = self.kv_cache_manager.take_events()
         if events:

vllm/v1/worker/gpu_model_runner.py

@@ -1747,7 +1747,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         lmcache_connector = get_lmcache_connector()
         if lmcache_connector is not None:
             lmcache_connector.bind_connector_metadata(
-                scheduler_output.kv_connector_metadata)
+                scheduler_output.kv_lmcache_connector_metadata)
             lmcache_connector.start_load_kv(get_forward_context())
 
     @staticmethod