[model-gateway]: add gRPC router embeddings endpoint implementation (#15273)

python/sglang/srt/entrypoints/grpc_server.py

@@ -455,10 +455,22 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer)
         input_text = grpc_req.tokenized.original_text
         input_ids = list(grpc_req.tokenized.input_ids)
 
+        # Convert sampling params
+        sampling_params = self._convert_sampling_params(grpc_req.sampling_params)
+
+        # For embedding requests, max_new_tokens should be 0.
+        # The scheduler logic expects an integer, not None.
+        sampling_params.max_new_tokens = 0
+
+        sampling_params.normalize(tokenizer=None)
+
         return TokenizedEmbeddingReqInput(
             rid=grpc_req.request_id,
             input_text=input_text,
             input_ids=input_ids,
+            image_inputs={"mm_items": []},
+            token_type_ids=list(grpc_req.token_type_ids),
+            sampling_params=sampling_params,
         )
 
     def _convert_sampling_params(