diff --git a/python/sglang/srt/disaggregation/common/conn.py b/python/sglang/srt/disaggregation/common/conn.py
index 595bb3ef1..ef0ba5245 100644
--- a/python/sglang/srt/disaggregation/common/conn.py
+++ b/python/sglang/srt/disaggregation/common/conn.py
@@ -164,6 +164,17 @@ class CommonKVManager(BaseKVManager):
         if num_kv_layers == dst_num_total_layers:
             dst_k_ptrs = dst_kv_ptrs[:dst_num_total_layers]
             dst_v_ptrs = dst_kv_ptrs[dst_num_total_layers:]
+        elif (
+            num_kv_layers < dst_num_total_layers
+            and dst_num_total_layers % num_kv_layers != 0
+        ):
+            # Case: Decode has more layers than Prefill (e.g., Decode has draft model KV while Prefill is deployed without speculative decoding)
+            # To prevent empty Value Cache, which leads to wrong response
+            # dst_kv_ptrs layout: [K_main..., V_main..., draft_K..., draft_V...]
+            dst_k_ptrs = dst_kv_ptrs[start_layer:end_layer]
+            dst_v_ptrs = dst_kv_ptrs[
+                num_kv_layers + start_layer : num_kv_layers + end_layer
+            ]
         else:
             # Decode pp size should be equal to prefill pp size or 1
             dst_k_ptrs = dst_kv_ptrs[start_layer:end_layer]