diff --git a/vllm/model_executor/models/interfaces.py b/vllm/model_executor/models/interfaces.py
index 7caf4c2207..d278af0607 100644
--- a/vllm/model_executor/models/interfaces.py
+++ b/vllm/model_executor/models/interfaces.py
@@ -207,7 +207,8 @@ class SupportsMultiModal(Protocol):
 
         raise NotImplementedError(
             f"No language model found in {type(self).__name__}! "
-            "You should initialize it via `_mark_language_model`."
+            "You should initialize it via `_mark_language_model`, "
+            "and make sure `embed_input_ids` is implemented."
         )
 
     @contextmanager
diff --git a/vllm/model_executor/models/nemotron_parse.py b/vllm/model_executor/models/nemotron_parse.py
index ae417f095e..6e52e7cfe7 100644
--- a/vllm/model_executor/models/nemotron_parse.py
+++ b/vllm/model_executor/models/nemotron_parse.py
@@ -281,6 +281,9 @@ class MBartDecoderNoPos(nn.Module):
         self.layernorm_embedding = nn.LayerNorm(config.d_model)
         self.layer_norm = nn.LayerNorm(config.d_model)
 
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.embed_tokens(input_ids)
+
     def forward(
         self,
         decoder_input_ids: torch.Tensor | None,