Add Support for Qwen2-VL Multi-modal Embedding Models (#3694)

python/sglang/srt/entrypoints/engine.py

@@ -214,13 +214,13 @@ class Engine:
     def encode(
         self,
         prompt: Union[str, List[str], List[Dict], List[List[Dict]]],
+        image_data: Optional[Union[List[str], str]] = None,
     ) -> Dict:
         """
         The arguments of this function is the same as `sglang/srt/managers/io_struct.py::EmbeddingReqInput`.
         Please refer to `EmbeddingReqInput` for the documentation.
         """
-
-        obj = EmbeddingReqInput(text=prompt)
+        obj = EmbeddingReqInput(text=prompt, image_data=image_data)
         loop = asyncio.get_event_loop()
         generator = self.tokenizer_manager.generate_request(obj, None)
         ret = loop.run_until_complete(generator.__anext__())