[Cleanup] Remove unused `KVConnectorModelRunnerMixin` methods (#32077)

Signed-off-by: Nick Hill <nickhill123@gmail.com>

tests/v1/kv_connector/unit/test_nixl_connector.py

@@ -518,7 +518,7 @@ class TestNixlHandshake:
                 )
             connector.bind_connector_metadata(metadata)
 
-            # Mimic maybe_setup_kv_connector in gpu_model_runner.
+            # Mimic logic in KVConnectorModelRunnerMixin._get_kv_connector_output.
             dummy_ctx = ForwardContext(
                 no_compile_layers={},
                 attn_metadata={},
@@ -531,7 +531,7 @@ class TestNixlHandshake:
                 f"start_load_kv took {_after_load - _before_load} seconds"
             )
 
-            # Mimic get_finished_kv_transfers in gpu_model_runner.
+            # Mimic logic in KVConnectorModelRunnerMixin._get_kv_connector_output.
             _, done_recving = connector.get_finished(finished_req_ids=set())
             if len(done_recving) > 0:
                 assert request_id in done_recving

vllm/v1/worker/kv_connector_model_runner_mixin.py

@@ -38,42 +38,12 @@ logger = init_logger(__name__)
 
 # Defined as a kv connector functionality mixin for ModelRunner (GPU, TPU)
 class KVConnectorModelRunnerMixin:
-    @staticmethod
-    def maybe_setup_kv_connector(scheduler_output: "SchedulerOutput"):
-        # Update KVConnector with the KVConnector metadata forward().
-        if has_kv_transfer_group():
-            kv_connector = get_kv_transfer_group()
-            assert isinstance(kv_connector, KVConnectorBase)
-            assert scheduler_output.kv_connector_metadata is not None
-            kv_connector.bind_connector_metadata(scheduler_output.kv_connector_metadata)
-
-            # Background KV cache transfers happen here.
-            # These transfers are designed to be async and the requests
-            # involved may be disjoint from the running requests.
-            # Do this here to save a collective_rpc.
-            kv_connector.start_load_kv(get_forward_context())
-
     @staticmethod
     def ensure_kv_transfer_shutdown() -> None:
         # has_kv_transfer_group can be None during interpreter shutdown.
         if has_kv_transfer_group and has_kv_transfer_group():  # type: ignore[truthy-function]
             ensure_kv_transfer_shutdown()
 
-    @staticmethod
-    def maybe_wait_for_kv_save() -> None:
-        if has_kv_transfer_group():
-            get_kv_transfer_group().wait_for_save()
-
-    @staticmethod
-    def get_finished_kv_transfers(
-        scheduler_output: "SchedulerOutput",
-    ) -> tuple[set[str] | None, set[str] | None]:
-        if has_kv_transfer_group():
-            return get_kv_transfer_group().get_finished(
-                scheduler_output.finished_req_ids
-            )
-        return None, None
-
     @staticmethod
     def kv_connector_no_forward(
         scheduler_output: "SchedulerOutput", vllm_config: VllmConfig