fix bug in case of no kvconnectorgroup

Signed-off-by: Will Eaton <weaton@redhat.com>

Author: wseaton

vllm/v1/worker/gpu_worker.py

@@ -18,6 +18,7 @@
                               set_custom_all_reduce)
 from vllm.distributed.kv_transfer import (ensure_kv_transfer_initialized,
                                           get_kv_transfer_group,
+                                          has_kv_transfer_group,
                                           is_v1_kv_transfer_group)
 from vllm.distributed.parallel_state import get_pp_group, get_tp_group
 from vllm.logger import init_logger
@@ -236,6 +237,9 @@ def determine_available_memory(self) -> int:
     def get_kv_connector_handshake_metadata(self) -> Optional[dict]:
         """Get KV connector metadata from this worker if available."""
 
+        if not has_kv_transfer_group():
+            return None
+
         connector = get_kv_transfer_group()
         if not is_v1_kv_transfer_group(connector):
             logger.warning("The KV connector is not a v1 connector. "