DOC: TPU mention fix (#38129)

Signed-off-by: Mateusz Sokół <mat646@gmail.com>

vllm/config/parallel.py

@@ -214,14 +214,18 @@ class ParallelConfig:
     distributed_executor_backend: (
         str | DistributedExecutorBackend | type[Executor] | None
     ) = None
-    """Backend to use for distributed model workers, either "ray" or "mp"
+    """
+    Backend to use for distributed model workers, either "ray" or "mp"
     (multiprocessing). If the product of pipeline_parallel_size and tensor_parallel_size
     is less than or equal to the number of GPUs available, "mp" will be used to
     keep processing on a single host. Otherwise, an error will be raised. To use "mp"
     you must also set nnodes, and to use "ray" you must manually set
     distributed_executor_backend to "ray".
 
-    Note that tpu only support Ray for distributed inference."""
+    Note:
+        [TPU](https://docs.vllm.ai/projects/tpu/en/latest/) platform only supports Ray
+        for distributed inference.
+    """
 
     worker_cls: str = "auto"
     """The full name of the worker class to use. If "auto", the worker class