[Bugfix][TPU] Fix TPU sampler output (vllm-project#5978)

vllm/worker/tpu_worker.py

@@ -215,7 +215,7 @@ def execute_model(
         assert len(seq_group_metadata_list) > 0
         output = self.model_runner.execute_model(seq_group_metadata_list,
                                                  self.tpu_cache)
-        return [output]
+        return output
 
     def cache_swap(
         self,