[misc] Add bench_serving compatibility shim (#28997)

python/sglang/bench_offline_throughput.py

@@ -15,7 +15,7 @@ warnings.warn(
     "`sglang.bench_offline_throughput` is deprecated and will be removed in a "
     "future release; use `sglang.benchmark.offline_throughput` instead "
     "(e.g. `python -m sglang.benchmark.offline_throughput`).",
-    DeprecationWarning,
+    FutureWarning,
     stacklevel=1,
 )
 

python/sglang/bench_one_batch.py

@@ -14,7 +14,7 @@ warnings.warn(
     "`sglang.bench_one_batch` is deprecated and will be removed in a future "
     "release; use `sglang.benchmark.one_batch` instead "
     "(e.g. `python -m sglang.benchmark.one_batch`).",
-    DeprecationWarning,
+    FutureWarning,
     stacklevel=1,
 )
 

python/sglang/bench_one_batch_server.py

@@ -4,8 +4,18 @@
 ``from sglang.bench_one_batch_server import ...`` imports.
 """
 
+import warnings
+
 from sglang.benchmark.one_batch_server import *  # noqa: F401,F403
 from sglang.benchmark.one_batch_server import main
 
+warnings.warn(
+    "`sglang.bench_one_batch_server` is deprecated and will be removed in a "
+    "future release; use `sglang.benchmark.one_batch_server` instead "
+    "(e.g. `python -m sglang.benchmark.one_batch_server`).",
+    FutureWarning,
+    stacklevel=1,
+)
+
 if __name__ == "__main__":
     main()

python/sglang/bench_serving.py

@@ -0,0 +1,20 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Compatibility shim for the relocated serving benchmark entrypoint."""
+
+import warnings
+
+from sglang.benchmark.serving import *  # noqa: F403
+from sglang.benchmark.serving import (  # noqa: F401
+    _create_bench_client_session,
+    cli_main,
+)
+
+warnings.warn(
+    "sglang.bench_serving is deprecated; use sglang.benchmark.serving instead.",
+    FutureWarning,
+    stacklevel=1,
+)
+
+
+if __name__ == "__main__":
+    cli_main()

python/sglang/benchmark/serving.py

@@ -7,9 +7,9 @@
 Benchmark online serving with dynamic requests.
 
 Usage:
-python3 -m sglang.bench_serving --backend sglang --num-prompt 10
+python3 -m sglang.benchmark.serving --backend sglang --num-prompt 10
 
-python3 -m sglang.bench_serving --backend sglang --dataset-name random --num-prompts 3000 --random-input 1024 --random-output 1024 --random-range-ratio 0.5
+python3 -m sglang.benchmark.serving --backend sglang --dataset-name random --num-prompts 3000 --random-input 1024 --random-output 1024 --random-range-ratio 0.5
 """
 
 import argparse
@@ -2124,7 +2124,7 @@ class LoRAPathAction(argparse.Action):
             getattr(namespace, self.dest).append(lora_name)
 
 
-if __name__ == "__main__":
+def cli_main():
     parser = ArgumentParser(description="Benchmark the online serving throughput.")
     parser.add_argument(
         "--backend",
@@ -2651,3 +2651,7 @@ if __name__ == "__main__":
     args = parser.parse_args()
     _validate_parsed_gsp_args(parser, args)
     run_benchmark(args)
+
+
+if __name__ == "__main__":
+    cli_main()