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, 01:00 UTC
[Feature][eplb] add verify ep or tp or dp (#21102)
Signed-off-by: rongfu.leng <rongfu.leng@daocloud.io>
rongfu.leng committed
Jul 22, 2025 at 14:41 UTC
3779eb8c81449b924a23457fc77e45a0e6171178
1 file changed
+9
vllm/config.py
+9
@@ -2108,6 +2108,15 @@ class ParallelConfig:
2108
raise ValueError(
2109
"num_redundant_experts must be non-negative, but got "
2110
f"{self.num_redundant_experts}.")
2111
+ if not self.enable_expert_parallel:
2112
+ raise ValueError(
2113
+ "enable_expert_parallel must be True to use EPLB.")
2114
+ if self.tensor_parallel_size * self.data_parallel_size <= 1:
2115
+ raise ValueError(
2116
+ "EPLB requires tensor_parallel_size or data_parallel_size "
2117
+ f"to be greater than 1, but got "
2118
+ f"TP={self.tensor_parallel_size},DP={self.data_parallel_size}."
2119
+ )
2120
else:
2121
if self.num_redundant_experts != 0:
2122
raise ValueError(