From 8278825b577675fa5fabfc3a38a48d55296a55b9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Mateusz=20Sok=C3=B3=C5=82?= <8431159+mtsokol@users.noreply.github.com> Date: Tue, 31 Mar 2026 12:27:56 +0200 Subject: [PATCH] DOC: TPU mention fix (#38129) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Mateusz Sokół --- vllm/config/parallel.py | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/vllm/config/parallel.py b/vllm/config/parallel.py index cab415a78f..339c912590 100644 --- a/vllm/config/parallel.py +++ b/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