add sglang version to get_server_info (#2206)

python/sglang/srt/server.py

@@ -86,6 +86,7 @@ from sglang.srt.utils import (
     set_ulimit,
 )
 from sglang.utils import get_exception_traceback
+from sglang.version import __version__
 
 logger = logging.getLogger(__name__)
 
@@ -527,6 +528,7 @@ async def _get_server_info():
         **dataclasses.asdict(tokenizer_manager.server_args),  # server args
         "memory_pool_size": await tokenizer_manager.get_memory_pool_size(),  # memory pool size
         "max_total_num_tokens": _max_total_num_tokens,  # max total num tokens
+        "version": __version__,
     }
 
 

test/srt/test_srt_endpoint.py

@@ -226,6 +226,9 @@ class TestSRTEndpoint(unittest.TestCase):
         attention_backend = response_json["attention_backend"]
         self.assertIsInstance(attention_backend, str)
 
+        version = response_json["version"]
+        self.assertIsInstance(version, str)
+
 
 if __name__ == "__main__":
     unittest.main()