From 4708f13a9c8758d08cd45df200d63aa97762e873 Mon Sep 17 00:00:00 2001 From: Michael Goin Date: Fri, 4 Apr 2025 11:58:08 -0600 Subject: [PATCH] [Bugfix] Fix default behavior/fallback for pp in v1 (#16057) Signed-off-by: mgoin --- vllm/engine/arg_utils.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py index 88723d9f5b741..89c9b67470e6b 100644 --- a/vllm/engine/arg_utils.py +++ b/vllm/engine/arg_utils.py @@ -1521,8 +1521,9 @@ class EngineArgs: # PP is supported on V1 with Ray distributed executor, # but off for MP distributed executor for now. if (self.pipeline_parallel_size > 1 - and self.distributed_executor_backend == "mp" - and _warn_or_fallback("PP (MP distributed executor)")): + and self.distributed_executor_backend != "ray"): + name = "Pipeline Parallelism without Ray distributed executor" + _raise_or_fallback(feature_name=name, recommend_to_remove=False) return False # ngram is supported on V1, but off by default for now.