[Misc] Avoid redundant copy for encoder-only models (#24012)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

vllm/v1/worker/gpu_model_runner.py

@@ -827,13 +827,13 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
                 blk_table_tensor = torch.zeros(
                     (num_reqs, 1),
                     dtype=torch.int32,
-                    pin_memory=self.pin_memory,
-                    device="cpu").to(self.device, non_blocking=True)
-                slot_mapping = torch.zeros((total_num_scheduled_tokens, ),
-                                           dtype=torch.int32,
-                                           pin_memory=self.pin_memory,
-                                           device="cpu").to(self.device,
-                                                            non_blocking=True)
+                    device=self.device,
+                )
+                slot_mapping = torch.zeros(
+                    (total_num_scheduled_tokens, ),
+                    dtype=torch.int64,
+                    device=self.device,
+                )
                 num_common_prefix_blocks = 0
             else:
                 blk_table = self.input_batch.block_table[kv_cache_group_id]