From 969aec4bc845794581fcfa5f3c0bad42023b8358 Mon Sep 17 00:00:00 2001 From: Wei Zhao <51183510+wzhao18@users.noreply.github.com> Date: Tue, 2 Jun 2026 21:26:30 -0400 Subject: [PATCH] [Bugfix] Fix Deepseek v4 non-mega-moe model init error (#44356) Signed-off-by: wzhao18 --- vllm/models/deepseek_v4/nvidia/model.py | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/vllm/models/deepseek_v4/nvidia/model.py b/vllm/models/deepseek_v4/nvidia/model.py index bd1ccc4cfb..547048ab58 100644 --- a/vllm/models/deepseek_v4/nvidia/model.py +++ b/vllm/models/deepseek_v4/nvidia/model.py @@ -620,6 +620,14 @@ class DeepseekV4MoE(nn.Module): self.experts_start_idx = self.tp_rank * self.n_local_experts self.experts_end_idx = self.experts_start_idx + self.n_local_experts + self.n_redundant_experts = 0 + self.n_shared_experts = config.n_shared_experts or 0 + self.n_logical_experts = self.n_routed_experts + self.n_physical_experts = self.n_logical_experts + self.n_local_physical_experts = self.n_local_experts + self.physical_expert_start = self.experts_start_idx + self.physical_expert_end = self.experts_end_idx + self.experts = FusedMoE( shared_experts=self.shared_experts, gate=self.gate,