[Engine] Fix generate hanging issue after the first call (#1606)

python/sglang/srt/server.py

@@ -691,8 +691,9 @@ class Engine:
             lora_path=lora_path,
         )
 
-        # make it synchronous
-        return asyncio.run(generate_request(obj, None))
+        # get the current event loop
+        loop = asyncio.get_event_loop()
+        return loop.run_until_complete(generate_request(obj, None))
 
     def shutdown(self):
         kill_child_process(os.getpid(), including_parent=False)

test/srt/test_srt_engine.py

@@ -28,6 +28,18 @@ class TestSRTBackend(unittest.TestCase):
         print(out2)
         assert out1 == out2, f"{out1} != {out2}"
 
+    def test_engine_multiple_generate(self):
+        # just to ensure there is no issue running multiple generate calls
+        prompt = "Today is a sunny day and I like"
+        model_path = DEFAULT_MODEL_NAME_FOR_TEST
+
+        sampling_params = {"temperature": 0, "max_new_tokens": 8}
+
+        engine = sgl.Engine(model_path=model_path, random_seed=42)
+        engine.generate(prompt, sampling_params)
+        engine.generate(prompt, sampling_params)
+        engine.shutdown()
+
 
 if __name__ == "__main__":
     unittest.main()