Fix failed ci tests on long prompts; Better error messages for embedding models (#1700)

python/sglang/srt/managers/io_struct.py

@@ -56,6 +56,9 @@ class GenerateReqInput:
     # LoRA related
     lora_path: Optional[Union[List[Optional[str]], Optional[str]]] = None
 
+    # Whether it is a single request or a batch request
+    is_single: bool = True
+
     def post_init(self):
         if (self.text is None and self.input_ids is None) or (
             self.text is not None and self.input_ids is not None

python/sglang/srt/managers/tokenizer_manager.py

@@ -150,9 +150,13 @@ class TokenizerManager:
         while self.model_update_lock.locked():
             await asyncio.sleep(0.001)
 
+        if isinstance(obj, EmbeddingReqInput) and self.is_generation:
+            raise ValueError(
+                "This model does not appear to be an embedding model by default. Please add `--is-embedding` when launching the server or try another model."
+            )
+
         obj.post_init()
         is_single = obj.is_single
-
         if is_single:
             async for response in self._handle_single_request(obj, request):
                 yield response

python/sglang/srt/server.py

@@ -542,8 +542,6 @@ def _wait_and_warmup(server_args, pipe_finish_writer, pid):
         kill_child_process(pid, including_parent=False)
         return
 
-    print(f"{res.json()=}")
-
     logger.info("The server is fired up and ready to roll!")
     if pipe_finish_writer is not None:
         pipe_finish_writer.send("ready")