KVConnector interface used by GPUModelRunner.
Source code in vllm/v1/worker/gpu/kv_connector.py
| class KVConnector:
"""KVConnector interface used by GPUModelRunner."""
def pre_forward(self, scheduler_output: "SchedulerOutput", **kwargs: Any) -> None:
pass
def finish_forward(self) -> None:
pass
def post_forward(self, finished_req_ids: set[str]) -> KVConnectorOutput | None:
return None
def no_forward(self, scheduler_output: "SchedulerOutput") -> ModelRunnerOutput:
return EMPTY_MODEL_RUNNER_OUTPUT
def set_disabled(self, disabled: bool) -> None:
pass
def reset_capture_state(self) -> None:
pass
|