benchscope 1.0.0__tar.gz

benchscope-1.0.0/LICENSE

@@ -0,0 +1,176 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS

benchscope-1.0.0/PKG-INFO

@@ -0,0 +1,136 @@
+Metadata-Version: 2.4
+Name: benchscope
+Version: 1.0.0
+Summary: vLLM / SGLang 推理服务性能测试 Web 工具 (benchscope)
+Author-email: benchscope <benchscope@example.com>
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/benchscope/benchscope
+Project-URL: Documentation, https://github.com/benchscope/benchscope#readme
+Project-URL: Source, https://github.com/benchscope/benchscope
+Keywords: vllm,sglang,benchmark,llm,performance-test,inference,openai-api,web-ui,benchscope
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: System :: Benchmark
+Classifier: Framework :: FastAPI
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: fastapi>=0.110
+Requires-Dist: uvicorn[standard]>=0.29
+Requires-Dist: requests>=2.31
+Requires-Dist: openpyxl>=3.1
+Requires-Dist: pydantic>=2
+Requires-Dist: python-multipart>=0.0.9
+Provides-Extra: modelscope
+Requires-Dist: modelscope>=1.15; extra == "modelscope"
+Dynamic: license-file
+
+# benchscope
+
+vLLM / SGLang 推理服务性能测试 Web 工具（类似 tensorboard-dashboard 的单进程 pip 工具）。
+
+连接 vLLM / SGLang（或任意 OpenAI 兼容）推理服务的 API 即可进行性能压测，
+实时展示并发数-吞吐/时延曲线、日志管理与 xlsx 汇总分析。
+
+## 功能特性（V1）
+
+- **双框架**：vLLM（`vllm bench serve`）与 SGLang（`sglang.bench_serving`），
+  bench 工具在 benchscope 所在机器以子进程运行，**推理服务端无需安装任何插件**，只需提供 OpenAI 兼容 API。
+- **数据集**：
+  - `random`：可多选输入/输出长度组合（默认 3K/1K、1K/1K、256/256，可自定义）；
+  - `sharegpt`：自动从 modelscope 下载
+    [gliang1001/ShareGPT_V3_unfiltered_cleaned_split](https://www.modelscope.cn/datasets/gliang1001/ShareGPT_V3_unfiltered_cleaned_split)
+    （JSON 数组自动流式转换为 jsonl 缓存）；
+  - `自定义数据集`：支持网页上传 jsonl 或填写服务器本地路径，功能与 sharegpt 一致。
+- **测试配置**：并发数列表可编辑（默认 1,4,8,16,32,40,64,128），
+  `--max-concurrency` 与 `--num-prompts` 保持一致，请求速率可选 `inf`（不限速）；
+  GPU 型号/数量自动检测（nvidia-smi）并可手动编辑；TPOT 阈值设置与最佳并发高亮；
+  框架常用参数表单 + 自由参数编辑器。
+- **实时结果**：每个并发结果实时入表（表头中英双语），
+  六条曲线（Output 吞吐 / Total 吞吐 / TTFT mean / TPOT mean / TTFT P99 / TPOT P99），横轴并发数。
+- **日志**：每次测试一个 `月日-时分秒` 目录，含原始 bench 日志、mean 汇总 CSV、P99 汇总 CSV、
+  `benchmark-*.xlsx`（均值 / P99 双 sheet，列：GPU、模型、精度、推理框架、输入长度、输出长度、
+  并发数、output、peakoutput、total、ttft、itl、tpot、单用户=1000/tpot），支持预览与下载。
+- **分析**：均值 / P99 两大分析块，output/peakoutput/total/ttft/itl/tpot 六曲线；
+  指定 TPOT 阈值后自动标记最接近（低于）阈值的记录为最佳并发。
+- **状态监控**：顶部导航实时显示 网站服务 / 推理服务 就绪或离线，推理服务断开自动变更状态。
+
+## 安装与启动
+
+```bash
+# 后端依赖
+pip install -e .
+
+# 前端（可选，若 webui/ 已有构建产物可跳过）
+cd web && npm install && npm run build && cd ..
+
+# 启动（默认 http://0.0.0.0:8080，自动打开浏览器）
+benchscope
+# 或指定端口 / 不自动开浏览器
+benchscope --port 8080 --no-browser
+```
+
+开发模式：
+
+```bash
+# 终端 1：后端
+python -m benchscope.cli --port 8080 --no-browser
+# 终端 2：前端热更新（代理到 8080）
+cd web && npm run dev    # http://127.0.0.1:5173
+```
+
+## 服务设置
+
+打开页面右上角「服务设置」：
+
+- Base URL：推理服务 OpenAI 兼容地址（默认 `http://192.168.1.67:8000`），
+  Endpoint、API Key、额外请求头；
+- GPU 自动检测（nvidia-smi）或手动填写；
+- 日志目录 `logs_dir` 与数据集缓存目录 `datasets_dir`（默认 `./logs`、`./datasets`）；
+- TPOT 阈值默认值、请求速率默认值；
+- bench 执行命令模板：`vllm bench serve` / `python -m sglang.bench_serving`。
+
+## 使用流程
+
+1. 打开页面，确认顶部「推理服务：就绪」；
+2. 进入 vLLM 或 SGLang 测试页，选择模型（来自 `/v1/models`）、数据集、并发数等；
+3. 点击「命令预览」确认将要执行的 bench 命令，点击「开始测试」；
+4. 实时结果面板逐条刷新，曲线实时更新；日志分析面板可查看原始日志、汇总与 xlsx；
+5. 测试完成后自动生成 `logs/<月日-时分秒>/benchmark-*.xlsx`（均值 + P99 双 sheet）。
+
+## 目录结构
+
+```
+benchscope/
+├── benchscope/
+│   ├── cli.py                # benchscope 命令入口
+│   ├── config.py             # 配置持久化 (~/.benchscope/config.json)
+│   ├── constants.py          # 默认值（并发列表、长度组合等）
+│   ├── datasets.py           # sharegpt 下载/转换、自定义数据集
+│   ├── gpu.py                # GPU 自动检测
+│   ├── parser.py             # bench 输出解析（mean + P99）
+│   ├── summary.py            # CSV 与 xlsx 汇总生成
+│   ├── benches/              # vllm/sglang 命令构建与执行
+│   └── server/               # FastAPI + WebSocket + 测试编排
+├── web/                      # Vue3 + Ant Design Vue 前端源码
+├── asserts/                  # 参考脚本与样例日志（test-non.py 等）
+└── tests/mock_openai_server.py  # 本地联调用模拟 OpenAI 服务
+```
+
+## 开发调试
+
+- 无 vllm/sglang 环境联调：设置环境变量 `BENCHSCOPE_FAKE_BENCH=1` 后启动，bench 将以仿真数据执行；
+- 本地模拟推理服务：`python tests/mock_openai_server.py`（端口 8001），
+  在服务设置中把 Base URL 指向 `http://127.0.0.1:8001`。
+
+## 规划
+
+V1 纯文本性能测试（本版本）→ V2 多模态模型测试 → V3 全模态 → V4 世界模型；
+V5 常见数据集精度测试 → V6 modelscope 官方模型对比分析。

benchscope-1.0.0/README.md

@@ -0,0 +1,102 @@
+# benchscope
+
+vLLM / SGLang 推理服务性能测试 Web 工具（类似 tensorboard-dashboard 的单进程 pip 工具）。
+
+连接 vLLM / SGLang（或任意 OpenAI 兼容）推理服务的 API 即可进行性能压测，
+实时展示并发数-吞吐/时延曲线、日志管理与 xlsx 汇总分析。
+
+## 功能特性（V1）
+
+- **双框架**：vLLM（`vllm bench serve`）与 SGLang（`sglang.bench_serving`），
+  bench 工具在 benchscope 所在机器以子进程运行，**推理服务端无需安装任何插件**，只需提供 OpenAI 兼容 API。
+- **数据集**：
+  - `random`：可多选输入/输出长度组合（默认 3K/1K、1K/1K、256/256，可自定义）；
+  - `sharegpt`：自动从 modelscope 下载
+    [gliang1001/ShareGPT_V3_unfiltered_cleaned_split](https://www.modelscope.cn/datasets/gliang1001/ShareGPT_V3_unfiltered_cleaned_split)
+    （JSON 数组自动流式转换为 jsonl 缓存）；
+  - `自定义数据集`：支持网页上传 jsonl 或填写服务器本地路径，功能与 sharegpt 一致。
+- **测试配置**：并发数列表可编辑（默认 1,4,8,16,32,40,64,128），
+  `--max-concurrency` 与 `--num-prompts` 保持一致，请求速率可选 `inf`（不限速）；
+  GPU 型号/数量自动检测（nvidia-smi）并可手动编辑；TPOT 阈值设置与最佳并发高亮；
+  框架常用参数表单 + 自由参数编辑器。
+- **实时结果**：每个并发结果实时入表（表头中英双语），
+  六条曲线（Output 吞吐 / Total 吞吐 / TTFT mean / TPOT mean / TTFT P99 / TPOT P99），横轴并发数。
+- **日志**：每次测试一个 `月日-时分秒` 目录，含原始 bench 日志、mean 汇总 CSV、P99 汇总 CSV、
+  `benchmark-*.xlsx`（均值 / P99 双 sheet，列：GPU、模型、精度、推理框架、输入长度、输出长度、
+  并发数、output、peakoutput、total、ttft、itl、tpot、单用户=1000/tpot），支持预览与下载。
+- **分析**：均值 / P99 两大分析块，output/peakoutput/total/ttft/itl/tpot 六曲线；
+  指定 TPOT 阈值后自动标记最接近（低于）阈值的记录为最佳并发。
+- **状态监控**：顶部导航实时显示 网站服务 / 推理服务 就绪或离线，推理服务断开自动变更状态。
+
+## 安装与启动
+
+```bash
+# 后端依赖
+pip install -e .
+
+# 前端（可选，若 webui/ 已有构建产物可跳过）
+cd web && npm install && npm run build && cd ..
+
+# 启动（默认 http://0.0.0.0:8080，自动打开浏览器）
+benchscope
+# 或指定端口 / 不自动开浏览器
+benchscope --port 8080 --no-browser
+```
+
+开发模式：
+
+```bash
+# 终端 1：后端
+python -m benchscope.cli --port 8080 --no-browser
+# 终端 2：前端热更新（代理到 8080）
+cd web && npm run dev    # http://127.0.0.1:5173
+```
+
+## 服务设置
+
+打开页面右上角「服务设置」：
+
+- Base URL：推理服务 OpenAI 兼容地址（默认 `http://192.168.1.67:8000`），
+  Endpoint、API Key、额外请求头；
+- GPU 自动检测（nvidia-smi）或手动填写；
+- 日志目录 `logs_dir` 与数据集缓存目录 `datasets_dir`（默认 `./logs`、`./datasets`）；
+- TPOT 阈值默认值、请求速率默认值；
+- bench 执行命令模板：`vllm bench serve` / `python -m sglang.bench_serving`。
+
+## 使用流程
+
+1. 打开页面，确认顶部「推理服务：就绪」；
+2. 进入 vLLM 或 SGLang 测试页，选择模型（来自 `/v1/models`）、数据集、并发数等；
+3. 点击「命令预览」确认将要执行的 bench 命令，点击「开始测试」；
+4. 实时结果面板逐条刷新，曲线实时更新；日志分析面板可查看原始日志、汇总与 xlsx；
+5. 测试完成后自动生成 `logs/<月日-时分秒>/benchmark-*.xlsx`（均值 + P99 双 sheet）。
+
+## 目录结构
+
+```
+benchscope/
+├── benchscope/
+│   ├── cli.py                # benchscope 命令入口
+│   ├── config.py             # 配置持久化 (~/.benchscope/config.json)
+│   ├── constants.py          # 默认值（并发列表、长度组合等）
+│   ├── datasets.py           # sharegpt 下载/转换、自定义数据集
+│   ├── gpu.py                # GPU 自动检测
+│   ├── parser.py             # bench 输出解析（mean + P99）
+│   ├── summary.py            # CSV 与 xlsx 汇总生成
+│   ├── benches/              # vllm/sglang 命令构建与执行
+│   └── server/               # FastAPI + WebSocket + 测试编排
+├── web/                      # Vue3 + Ant Design Vue 前端源码
+├── asserts/                  # 参考脚本与样例日志（test-non.py 等）
+└── tests/mock_openai_server.py  # 本地联调用模拟 OpenAI 服务
+```
+
+## 开发调试
+
+- 无 vllm/sglang 环境联调：设置环境变量 `BENCHSCOPE_FAKE_BENCH=1` 后启动，bench 将以仿真数据执行；
+- 本地模拟推理服务：`python tests/mock_openai_server.py`（端口 8001），
+  在服务设置中把 Base URL 指向 `http://127.0.0.1:8001`。
+
+## 规划
+
+V1 纯文本性能测试（本版本）→ V2 多模态模型测试 → V3 全模态 → V4 世界模型；
+V5 常见数据集精度测试 → V6 modelscope 官方模型对比分析。

benchscope-1.0.0/benchscope/__init__.py

@@ -0,0 +1,3 @@
+"""benchscope - vLLM / SGLang 推理服务性能测试 Web 工具."""
+
+__version__ = "1.0.0"

benchscope-1.0.0/benchscope/__main__.py

@@ -0,0 +1,4 @@
+from benchscope.cli import main
+
+if __name__ == "__main__":
+    main()

benchscope-1.0.0/benchscope/benches/__init__.py

@@ -0,0 +1 @@
+"""bench 命令构建与执行。"""

benchscope-1.0.0/benchscope/benches/base.py

@@ -0,0 +1,54 @@
+"""bench 命令构建的公共定义。"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+
+@dataclass
+class ParamDef:
+    """UI 表单中一个可配置参数的定义。"""
+
+    key: str          # 表单字段名
+    flag: str         # 实际 CLI flag，如 "--temperature"
+    label: str        # 中文标签
+    help: str = ""
+    type: str = "str"  # str | int | float | bool | select
+    default: Any = None
+    options: list = field(default_factory=list)
+    advanced: bool = False  # 是否归入“高级参数”折叠区
+
+
+@dataclass
+class BenchOptions:
+    """一次 bench 执行所需的全部选项。"""
+
+    framework: str
+    model: str
+    api: dict            # {host, port, base_url, endpoint, api_key, extra_headers}
+    dataset: dict        # {type, path, input_len, output_len, sharegpt_output_len}
+    concurrency: int
+    request_rate: str | float = "inf"
+    curated: dict = field(default_factory=dict)   # 表单参数 key -> value
+    extra_args: list = field(default_factory=list)  # [{"flag": "--x", "value": "y"}]
+
+
+def build_arg_list(flags: list[list]) -> list[str]:
+    """将 [["--flag","value"], ["--bool",""]] 展开为命令行列表。"""
+    out: list[str] = []
+    for item in flags:
+        flag, value = item[0], item[1] if len(item) > 1 else ""
+        if isinstance(value, bool):
+            if value:
+                out.append(flag)
+            continue
+        if value is None or value == "":
+            out.append(flag)
+        else:
+            out.extend([flag, str(value)])
+    return out
+
+
+def flag_value(flag: str, value: Any) -> list[str]:
+    """单个 flag 的展开（供参数校验后使用）。"""
+    return build_arg_list([[flag, value]])

benchscope-1.0.0/benchscope/benches/runner.py

@@ -0,0 +1,185 @@
+"""bench 子进程流式执行器。
+
+支持真实执行（vllm/sglang CLI）与 FAKE 模式（BENCHSCOPE_FAKE_BENCH=1，
+生成仿真输出，便于无 vllm/sglang 环境下联调 UI 全流程）。
+"""
+from __future__ import annotations
+
+import logging
+import math
+import os
+import random
+import shlex
+import subprocess
+import sys
+import threading
+import time
+from typing import Callable, Optional
+
+from benchscope.parser import parse_metrics
+
+log = logging.getLogger("benchscope.runner")
+
+StreamCallback = Callable[[str], None]  # 每行输出回调
+
+
+class StopRequested(RuntimeError):
+    """测试被人为停止。"""
+
+
+class BenchRunner:
+    def __init__(self, command_template: str | None = None):
+        """command_template 形如 "vllm bench serve" / "python -m sglang.bench_serving"。"""
+        self.command_template = command_template or "vllm bench serve"
+        self._proc: Optional[subprocess.Popen] = None
+        self._stop_flag = threading.Event()
+
+    def kill(self) -> None:
+        """终止当前执行的子进程（用于停止测试）。"""
+        self._stop_flag.set()
+        proc = self._proc
+        if proc and proc.poll() is None:
+            try:
+                proc.kill()
+            except Exception:
+                pass
+
+    # ------------------------------------------------------------------
+    def run(
+        self,
+        cmd: list[str],
+        stream_cb: Optional[StreamCallback] = None,
+        timeout: float | None = None,
+    ) -> dict:
+        """执行命令，返回 parse_metrics 结果（含 raw）。失败抛 RuntimeError。"""
+        self._stop_flag.clear()
+        if os.environ.get("BENCHSCOPE_FAKE_BENCH") == "1":
+            return self._run_fake(cmd, stream_cb)
+
+        # 用模板指定的可执行文件替换命令头部（vllm / python -m sglang...）
+        full_cmd = self._resolve(cmd)
+        log.info("执行命令: %s", " ".join(full_cmd))
+        if stream_cb:
+            stream_cb("$ " + " ".join(full_cmd) + "\n")
+
+        try:
+            proc = subprocess.Popen(
+                full_cmd,
+                stdout=subprocess.PIPE,
+                stderr=subprocess.STDOUT,
+                text=True,
+                encoding="utf-8",
+                errors="replace",
+                bufsize=1,
+            )
+            self._proc = proc
+        except FileNotFoundError as e:
+            raise RuntimeError(
+                f"未找到命令执行环境：{full_cmd[0]}。请确认已安装 "
+                f"{self.command_template.split()[0]} 相关 CLI（并在服务设置中配置 bench 命令）。"
+            ) from e
+
+        chunks: list[str] = []
+        start = time.time()
+        try:
+            assert proc.stdout is not None
+            for line in proc.stdout:
+                chunks.append(line)
+                if stream_cb:
+                    stream_cb(line)
+                if timeout and time.time() - start > timeout:
+                    proc.kill()
+                    raise RuntimeError(f"bench 执行超时（>{timeout}s）")
+            proc.wait()
+        except KeyboardInterrupt:
+            proc.kill()
+            raise
+        finally:
+            self._proc = None
+        if proc.returncode != 0:
+            output = "".join(chunks)
+            if proc.returncode == -9 or proc.returncode == 137:
+                raise StopRequested("测试已被停止")
+            raise RuntimeError(
+                f"bench 命令执行失败（返回码 {proc.returncode}）。\n完整日志：\n{output[-4000:]}"
+            )
+        output = "".join(chunks)
+        metrics = parse_metrics(output)
+        if "output_mean" not in metrics:
+            raise RuntimeError(f"未能从 bench 输出解析出指标，输出如下：\n{output[-3000:]}")
+        return metrics
+
+    # ------------------------------------------------------------------
+    def _resolve(self, cmd: list[str]) -> list[str]:
+        """把命令头替换为模板指定的执行方式。"""
+        tmpl = shlex.split(self.command_template)
+        # 保留原始参数（从模板之后开始）
+        return tmpl + cmd[len(tmpl):] if cmd[:len(tmpl)] == tmpl else tmpl + cmd
+
+    # ------------------------------------------------------------------
+    # FAKE 模式：生成仿真 vllm 风格输出
+    def _run_fake(self, cmd: list[str], stream_cb: Optional[StreamCallback] = None) -> dict:
+        args = " ".join(cmd)
+        concurrency = 1
+        input_len, output_len = 1024, 1024
+        for i, tok in enumerate(cmd):
+            if tok == "--max-concurrency" and i + 1 < len(cmd):
+                concurrency = int(cmd[i + 1])
+            if tok == "--random-input-len" and i + 1 < len(cmd):
+                input_len = int(cmd[i + 1])
+            if tok == "--random-output-len" and i + 1 < len(cmd):
+                output_len = int(cmd[i + 1])
+        rng = random.Random(int(time.time() * 1000) % 2**31)
+
+        c = max(concurrency, 1)
+        out_tps = round(50 * c**0.62 * rng.uniform(0.95, 1.05), 2)
+        total = round(out_tps * (input_len + output_len) / output_len, 2)
+        ttft = round(60 + 9 * c + rng.uniform(0, 20), 2)
+        tpot = round(18 + 0.55 * c + rng.uniform(0, 3), 2)
+        itl = round(tpot * rng.uniform(0.97, 1.02), 2)
+
+        lines = [
+            "============ Serving Benchmark Result ============",
+            "Successful requests:                     %d" % c,
+            "Failed requests:                         0",
+            "Maximum request concurrency:             %d" % c,
+            "Benchmark duration (s):                  %.2f" % rng.uniform(5, 40),
+            "Total input tokens:                      %d" % (input_len * c),
+            "Total generated tokens:                  %d" % (output_len * c),
+            "Request throughput (req/s):              %.2f" % rng.uniform(0.1, c),
+            "Output token throughput (tok/s):         %s" % out_tps,
+            "Peak output token throughput (tok/s):    %s" % round(out_tps * 1.02, 2),
+            "Peak concurrent requests:                %.2f" % c,
+            "Total token throughput (tok/s):          %s" % total,
+            "---------------Time to First Token----------------",
+            "Mean TTFT (ms):                          %s" % ttft,
+            "Median TTFT (ms):                        %s" % round(ttft * 0.98, 2),
+            "P99 TTFT (ms):                           %s" % round(ttft * rng.uniform(1.05, 1.3), 2),
+            "-----Time per Output Token (excl. 1st token)------",
+            "Mean TPOT (ms):                          %s" % tpot,
+            "Median TPOT (ms):                        %s" % round(tpot * 0.97, 2),
+            "P99 TPOT (ms):                           %s" % round(tpot * rng.uniform(1.06, 1.35), 2),
+            "---------------Inter-token Latency----------------",
+            "Mean ITL (ms):                           %s" % itl,
+            "Median ITL (ms):                         %s" % round(itl * 0.97, 2),
+            "P99 ITL (ms):                            %s" % round(itl * rng.uniform(1.05, 1.3), 2),
+            "==================================================",
+            "",
+        ]
+        output = "\n".join(lines)
+        # 模拟耗时（可被 kill 中断）
+        total_sleep = min(0.6, 0.2 + c * 0.01)
+        slept = 0.0
+        while slept < total_sleep:
+            if self._stop_flag.is_set():
+                raise StopRequested("测试已被停止")
+            time.sleep(0.05)
+            slept += 0.05
+        if self._stop_flag.is_set():
+            raise StopRequested("测试已被停止")
+        if stream_cb:
+            stream_cb(f"$ {args}\n")
+            for ln in lines:
+                stream_cb(ln + "\n")
+        metrics = parse_metrics(output)
+        return metrics