From 87e47eb1db3dc25e3508c09f119ed5d7ffd506d8 Mon Sep 17 00:00:00 2001 From: Tyler Michael Smith Date: Mon, 7 Apr 2025 19:56:41 +0000 Subject: [PATCH] Fix use_ep Signed-off-by: Tyler Michael Smith --- vllm/model_executor/layers/fused_moe/layer.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/model_executor/layers/fused_moe/layer.py b/vllm/model_executor/layers/fused_moe/layer.py index 0e35d8a80988c..fea13c2b43839 100644 --- a/vllm/model_executor/layers/fused_moe/layer.py +++ b/vllm/model_executor/layers/fused_moe/layer.py @@ -437,7 +437,7 @@ class FusedMoE(torch.nn.Module): # Use expert parallelism instead of tensor parallelism? vllm_config = get_current_vllm_config() use_ep = (vllm_config.parallel_config.enable_expert_parallel - and self.tp_size > 1) + and (self.tp_size != 1 or self.dp_size != 1)) # For smuggling this layer into the fused moe custom op self.use_direct_call = self.dp_size == 1