Fix TP>1 in encoder-decoder models (#607)

vllm/worker/hpu_enc_dec_model_runner.py

@@ -168,6 +168,7 @@ def as_broadcastable_tensor_dict(self) -> Dict[str, Any]:
         tensor_dict = {
             "input_tokens": self.input_tokens,
             "input_positions": self.input_positions,
+            "multi_modal_kwargs": self.multi_modal_kwargs,
             "encoder_input_tokens": self.encoder_input_tokens,
             "encoder_input_positions": self.encoder_input_positions,
         }