diff --git a/vllm/distributed/device_communicators/base_device_communicator.py b/vllm/distributed/device_communicators/base_device_communicator.py index 1bc2d8e0281c7..06c1257a0f03b 100644 --- a/vllm/distributed/device_communicators/base_device_communicator.py +++ b/vllm/distributed/device_communicators/base_device_communicator.py @@ -59,6 +59,9 @@ class All2AllManagerBase: # usually the underlying implementation caches the handle # and reuse it for the same config. raise NotImplementedError + + def get_handles(self, kwargs): + raise NotImplementedError def dispatch(self, hidden_states: torch.Tensor, router_logits: torch.Tensor):