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 14, 05:30 UTC
[Bugfix] Improve EPLB config validation error message (#24524)
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Tyler Michael Smith committed
Sep 9, 2025 at 20:32 UTC
561f38dc3cfc54cca7710b033a86a315deb24111
1 file changed
+4
-2
vllm/config/parallel.py
+4
-2
@@ -368,8 +368,10 @@ class ParallelConfig:
368
else:
369
if self.eplb_config.num_redundant_experts != 0:
370
raise ValueError(
371
- "num_redundant_experts should be used with EPLB."
372
- f"{self.eplb_config.num_redundant_experts}.")
371
+ "num_redundant_experts is set to "
372
+ f"{self.eplb_config.num_redundant_experts} but EPLB is not "
373
+ "enabled. Either enable EPLB or unset "
374
+ "num_redundant_experts.")
375
if self.distributed_executor_backend is None and self.world_size > 1:
376
# We use multiprocessing by default if world_size fits on the
377
# current node and we aren't in a ray placement group.