diff --git a/vllm/patches/deepseek_v4.py b/vllm/patches/deepseek_v4.py
index 3f519bc4..a6d7aa16 100644
--- a/vllm/patches/deepseek_v4.py
+++ b/vllm/patches/deepseek_v4.py
@@ -1277,8 +1277,9 @@ class DeepseekV4Model(nn.Module):
             ".compressor.kv_proj.": ".compressor.wkv.",
             ".compressor.gate_proj.": ".compressor.gate.",
             # Shared expert projections (stacking into gate_up_proj)
-            ".shared_experts.gate_proj.": ".shared_experts.w1.",
-            ".shared_experts.up_proj.": ".shared_experts.w3.",
+            # Checkpoint has .shared_experts. but model has .ffn.shared_experts.
+            ".shared_experts.gate_proj.": ".ffn.shared_experts.w1.",
+            ".shared_experts.up_proj.": ".ffn.shared_experts.w3.",
             # modelopt uses mlp, vllm uses ffn internally
             ".mlp.": ".ffn.",
         }
@@ -2096,8 +2097,9 @@ def _make_deepseek_v4_weights_mapper(expert_dtype: str) -> WeightsMapper:
             ".compressor.kv_proj.": ".compressor.wkv.",
             ".compressor.gate_proj.": ".compressor.wgate.",
             # Shared expert projections (stacking into gate_up_proj)
-            ".shared_experts.gate_proj.": ".shared_experts.w1.",
-            ".shared_experts.up_proj.": ".shared_experts.w3.",
+            # Checkpoint has .shared_experts. but model has .ffn.shared_experts.
+            ".shared_experts.gate_proj.": ".ffn.shared_experts.w1.",
+            ".shared_experts.up_proj.": ".ffn.shared_experts.w3.",
             # modelopt uses mlp, vllm uses ffn internally
             ".mlp.": ".ffn.",
         },