[misc] Move bench_serving into sglang.benchmark (#28996)

python/sglang/auto_benchmark_lib.py

@@ -1095,7 +1095,7 @@ def build_bench_command(
     command = [
         sys.executable,
         "-m",
-        "sglang.bench_serving",
+        "sglang.benchmark.serving",
         "--backend",
         backend,
         "--base-url",

python/sglang/test/ascend/test_ascend_utils.py

@@ -18,7 +18,7 @@ import subprocess
 from types import SimpleNamespace
 from typing import Awaitable, Callable, NamedTuple, Optional
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.srt.utils import kill_process_tree
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,

python/sglang/test/ci/ci_stress_utils.py

@@ -67,7 +67,7 @@ class StressTestRunner:
         command = [
             "python3",
             "-m",
-            "sglang.bench_serving",
+            "sglang.benchmark.serving",
             "--backend",
             "sglang-oai",
             "--base-url",

python/sglang/test/kits/cache_hit_kit.py

@@ -5,8 +5,8 @@ import time
 import aiohttp
 import requests
 
-from sglang.bench_serving import RequestFuncOutput
 from sglang.benchmark.datasets.random import sample_random_requests
+from sglang.benchmark.serving import RequestFuncOutput
 from sglang.benchmark.utils import get_tokenizer, remove_prefix
 
 AIOHTTP_TIMEOUT = aiohttp.ClientTimeout(total=20 * 60 * 60)

python/sglang/test/mock_model/utils.py

@@ -5,7 +5,7 @@ import subprocess
 from dataclasses import dataclass
 from typing import Any, Sequence
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.srt.utils import kill_process_tree
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,

python/sglang/test/test_utils.py

@@ -31,7 +31,7 @@ import torch
 import torch.nn.functional as F
 from PIL import Image
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.global_config import global_config
 from sglang.srt.environ import envs
 from sglang.srt.utils import (

test/manual/dsv4/test_dsv4_pro_mtp.py

@@ -22,7 +22,7 @@ import requests
 
 from sglang.bench_one_batch_server import BenchArgs as OneBatchBenchArgs
 from sglang.bench_one_batch_server import run_benchmark as run_one_batch_benchmark
-from sglang.bench_serving import run_benchmark as run_serving_benchmark
+from sglang.benchmark.serving import run_benchmark as run_serving_benchmark
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import kill_process_tree
 from sglang.test.test_utils import (

test/registered/bench_fn/test_bench_serving_functionality.py

@@ -6,7 +6,7 @@ import unittest
 from http.server import BaseHTTPRequestHandler, HTTPServer
 from pathlib import Path
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.benchmark.utils import parse_custom_headers
 from sglang.srt.constants import HEALTH_CHECK_RID_PREFIX
 from sglang.srt.utils import kill_process_tree

test/registered/bench_fn/test_bench_serving_reasoning_stream.py

@@ -16,7 +16,7 @@ import unittest
 from argparse import Namespace
 from http.server import BaseHTTPRequestHandler, HTTPServer
 
-from sglang.bench_serving import (
+from sglang.benchmark.serving import (
     RequestFuncInput,
     async_request_openai_chat_completions,
     calculate_metrics,

test/registered/bench_fn/test_benchmark_datasets_api.py

@@ -985,7 +985,7 @@ class TestBenchmarkDatasetsAPI(unittest.TestCase):
         # flags with the rank-based Zipf formula and the alpha constraint,
         # and argparse rejects an unknown distribution choice.
         help_res = subprocess.run(
-            [sys.executable, "-m", "sglang.bench_serving", "--help"],
+            [sys.executable, "-m", "sglang.benchmark.serving", "--help"],
             capture_output=True,
             text=True,
             timeout=90,
@@ -1005,7 +1005,7 @@ class TestBenchmarkDatasetsAPI(unittest.TestCase):
             [
                 sys.executable,
                 "-m",
-                "sglang.bench_serving",
+                "sglang.benchmark.serving",
                 "--dataset-name",
                 "generated-shared-prefix",
                 "--gsp-group-distribution",
@@ -1026,7 +1026,7 @@ class TestBenchmarkDatasetsAPI(unittest.TestCase):
             [
                 sys.executable,
                 "-m",
-                "sglang.bench_serving",
+                "sglang.benchmark.serving",
                 "--dataset-name",
                 "generated-shared-prefix",
                 "--gsp-group-distribution",
@@ -1061,7 +1061,7 @@ class TestBenchmarkDatasetsAPI(unittest.TestCase):
             [
                 sys.executable,
                 "-m",
-                "sglang.bench_serving",
+                "sglang.benchmark.serving",
                 "--dataset-name",
                 "generated-shared-prefix",
                 "--gsp-group-distribution",

test/registered/disaggregation/test_disaggregation_dp_attention.py

@@ -1,7 +1,7 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.srt.environ import envs
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval

test/registered/jit/test_deepseek_v4_compress_state_runtime_shapes.py

@@ -59,7 +59,7 @@
 #
 # Workload for the service-level scenario:
 #   HF_ENDPOINT=https://hf-mirror.com \
-#   python3 -m sglang.bench_serving \
+#   python3 -m sglang.benchmark.serving \
 #     --host localhost \
 #     --port 8080 \
 #     --model /data00/models/DeepSeek-V4-Flash \

test/registered/scheduler/test_prefill_delayer.py

@@ -11,7 +11,7 @@ import openai
 import requests
 import torch
 
-from sglang.bench_serving import run_benchmark
+from sglang.benchmark.serving import run_benchmark
 from sglang.srt.managers.prefill_delayer import PrefillDelayer
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci