A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 07:15 UTC
[Bugfix] Fix PP default fallback behavior for V1 (#18915)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
May 29, 2025 at 22:13 UTC
d54af615d5cf63e111e564b3a2c25405fc16b4fd
1 file changed
+2
-1
vllm/engine/arg_utils.py
+2
-1
@@ -1380,7 +1380,8 @@ class EngineArgs:
1380
1381
if (self.pipeline_parallel_size > 1
1382
and self.distributed_executor_backend
1383
- not in ("ray", "mp", "external_launcher")):
1383
+ not in (ParallelConfig.distributed_executor_backend, "ray",
1384
+ "mp", "external_launcher")):
1385
name = "Pipeline Parallelism without Ray distributed executor " \
1386
"or multiprocessing executor or external launcher"
1387
_raise_or_fallback(feature_name=name, recommend_to_remove=False)