[Bugfix] Fix for the condition to accept empty encoder inputs for mllama (vllm-project#17732)

Signed-off-by: Gregory Shtrasberg <[email protected]>
Signed-off-by: Mu Huai <[email protected]>

Author: gshtras

vllm/engine/llm_engine.py

@@ -2021,7 +2021,7 @@ def _validate_model_input(
         if not prompt_ids:
             if prompt_type == "encoder" and model_config.is_multimodal_model:
                 pass  # Mllama may have empty encoder inputs for text-only data
-            if prompt_inputs["type"] == "embeds":
+            elif prompt_inputs["type"] == "embeds":
                 pass
             else:
                 raise ValueError(f"The {prompt_type} prompt cannot be empty")