[Feature] modify Runtime to support skip_tokenizer_init (#1088)

Co-authored-by: lzhang <zhanglei@modelbest.cn>

python/sglang/srt/server.py

@@ -533,6 +533,13 @@ class Runtime:
         prompt: str,
         sampling_params: Optional[Dict] = None,
     ):
+        if self.server_args.skip_tokenizer_init:
+            json_data = {
+                "input_ids": prompt,
+                "sampling_params": sampling_params,
+                "stream": True,
+            }
+        else:
             json_data = {
                 "text": prompt,
                 "sampling_params": sampling_params,
@@ -549,10 +556,13 @@ class Runtime:
                         if chunk == "data: [DONE]\n\n":
                             break
                         data = json.loads(chunk[5:].strip("\n"))
+                        if hasattr(data, "text"):
                             cur = data["text"][pos:]
                             if cur:
                                 yield cur
                             pos += len(cur)
+                        else:
+                            yield data
 
     add_request = async_generate