SWE-bench_Verified_Agentic#
概述#
SWE-bench Verified Agentic 是对 SWE-bench Verified 的代理模式(agentic-mode)评估。SWE-bench Verified 是从 SWE-bench 中人工验证筛选出的 500 个样本子集。与提供 oracle 文件上下文的单轮变体不同,该模式要求模型在每个实例专属的 Docker 容器内,通过多轮代理循环自主探索代码仓库、执行 shell 命令、编辑源文件,并最终提交补丁。
任务描述#
任务类型:自动化软件工程 / 缺陷修复(代理模式)
输入:GitHub issue 描述(不包含 oracle 文件上下文)
输出:模型自主编辑后通过
git diff生成的代码补丁(diff 格式)涉及仓库:12 个流行的 Python 项目(如 Django、Flask、Requests 等)
主要特性#
包含 500 个人工验证的 Issue-Pull Request 对
支持多轮代理循环(兼容 mini-swe-agent 的
swebench.yaml配置)每个实例使用独立的 SWE-bench Docker 容器作为执行沙箱
基于哨兵值(sentinel)的补丁提交协议(
COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT)同时支持函数调用(
toolcall)和基于文本(backticks)的行动协议
评估说明#
评估前需安装
pip install swebench==4.1.0每个仓库的 Docker 镜像会自动构建或拉取
每个实例的最终补丁验证超时时间为 1800 秒(30 分钟)
详细设置说明请参阅 使用文档
支持本地构建镜像和远程拉取镜像两种方式
代理模式#
该基准测试在每个实例专属的 SWE-bench Docker 容器内驱动一个多轮代理循环(与 mini-swe-agent 的 swebench.yaml 配置一致)。模型通过发出 bash 命令来探索 /testbed 目录、编辑源文件,并最终通过打印哨兵值 COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT 及其后的补丁内容来提交 git diff 补丁。
extra_params.action_protocol 参数用于选择以下两种协议之一:
toolcall(默认):采用 OpenAI 函数调用协议,仅提供一个bash工具。推荐用于支持工具调用的模型。backticks:基于文本的备用协议,每轮期望一个```mswea_bash_command ```代码块。适用于不支持函数调用的模型。
属性#
属性 |
值 |
|---|---|
基准测试名称 |
|
数据集 ID |
|
论文 |
N/A |
标签 |
|
指标 |
|
默认示例数 |
0-shot |
评估分割 |
|
数据统计#
指标 |
值 |
|---|---|
总样本数 |
500 |
提示词长度(平均) |
1699.73 字符 |
提示词长度(最小/最大) |
143 / 24770 字符 |
样例示例#
子集: default
{
"input": [
{
"id": "360d65af",
"content": "Modeling's `separability_matrix` does not compute separability correctly for nested CompoundModels\nConsider the following model:\r\n\r\n```python\r\nfrom astropy.modeling import models as m\r\nfrom astropy.modeling.separable import separability_matri ... [TRUNCATED 762 chars] ... [ True, True, False, False],\r\n [False, False, True, True],\r\n [False, False, True, True]])\r\n```\r\nSuddenly the inputs and outputs are no longer separable?\r\n\r\nThis feels like a bug to me, but I might be missing something?\n"
}
],
"id": 0,
"group_id": 0,
"tools": [
{
"name": "bash",
"description": "Execute a bash command inside the sandbox environment. Returns the combined stdout / stderr output of the command.",
"parameters": {
"properties": {
"command": {
"type": "string",
"description": "The bash command to execute."
},
"timeout": {
"type": "number",
"description": "Maximum execution time in seconds (default: 60).",
"default": 60
}
},
"required": [
"command"
]
}
}
],
"metadata": {
"problem_statement": "Modeling's `separability_matrix` does not compute separability correctly for nested CompoundModels\nConsider the following model:\r\n\r\n```python\r\nfrom astropy.modeling import models as m\r\nfrom astropy.modeling.separable import separability_matri ... [TRUNCATED 762 chars] ... [ True, True, False, False],\r\n [False, False, True, True],\r\n [False, False, True, True]])\r\n```\r\nSuddenly the inputs and outputs are no longer separable?\r\n\r\nThis feels like a bug to me, but I might be missing something?\n",
"instance_id": "astropy__astropy-12907",
"base_commit": "d16bfe05a744909de4b27f5875fe0d4ed41ce607",
"patch": "diff --git a/astropy/modeling/separable.py b/astropy/modeling/separable.py\n--- a/astropy/modeling/separable.py\n+++ b/astropy/modeling/separable.py\n@@ -242,7 +242,7 @@ def _cstack(left, right):\n cright = _coord_matrix(right, 'right', noutp)\n else:\n cright = np.zeros((noutp, right.shape[1]))\n- cright[-right.shape[0]:, -right.shape[1]:] = 1\n+ cright[-right.shape[0]:, -right.shape[1]:] = right\n \n return np.hstack([cleft, cright])\n \n",
"PASS_TO_PASS": [
"astropy/modeling/tests/test_separable.py::test_coord_matrix",
"astropy/modeling/tests/test_separable.py::test_cdot",
"astropy/modeling/tests/test_separable.py::test_cstack",
"astropy/modeling/tests/test_separable.py::test_arith_oper",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model0-result0]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model1-result1]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model2-result2]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model3-result3]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model4-result4]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model5-result5]",
"... [TRUNCATED 3 more items] ..."
],
"FAIL_TO_PASS": [
"astropy/modeling/tests/test_separable.py::test_separable[compound_model6-result6]",
"astropy/modeling/tests/test_separable.py::test_separable[compound_model9-result9]"
],
"test_patch": "diff --git a/astropy/modeling/tests/test_separable.py b/astropy/modeling/tests/test_separable.py\n--- a/astropy/modeling/tests/test_separable.py\n+++ b/astropy/modeling/tests/test_separable.py\n@@ -28,6 +28,13 @@\n p1 = models.Polynomial1D(1, nam ... [TRUNCATED 931 chars] ... [True, True, False, False, False],\n+ [False, False, True, False, False],\n+ [False, False, False, True, False],\n+ [False, False, False, False, True]]))),\n }\n \n \n",
"version": "4.3",
"repo": "astropy/astropy",
"environment_setup_commit": "298ccb478e6bf092953bca67a3d29dc6c35f6752",
"hints_text": "",
"created_at": "2022-03-03T15:14:54Z",
"docker_image": "swebench/sweb.eval.arm64.astropy_1776_astropy-12907:latest"
}
}
提示模板#
提示模板:
{question}
额外参数#
参数 |
类型 |
默认值 |
描述 |
|---|---|---|---|
|
|
|
代理行动协议:"toolcall"(主流 OpenAI 函数调用方式,与 mini-swe-agent swebench.yaml 一致)或 "backticks"(针对不支持函数调用的模型的文本式 mswea_bash_command 回退方案)。可选值:['toolcall', 'backticks'] |
|
|
|
每个样本的最大代理步数。 |
|
|
|
每个 bash 命令的默认超时时间(秒)。 |
|
|
|
SWE-bench 容器内的工作目录。 |
|
|
|
为每个样本在本地构建 Docker 镜像。 |
|
|
|
在构建前尝试拉取已存在的远程 Docker 镜像。 |
|
|
`` |
可选地强制指定镜像构建/拉取的目标架构。可选值:['', 'arm64', 'x86_64'] |
使用方法#
使用 CLI#
evalscope eval \
--model YOUR_MODEL \
--api-url OPENAI_API_COMPAT_URL \
--api-key EMPTY_TOKEN \
--datasets swe_bench_verified_agentic \
--limit 10 # 正式评估时请删除此行
使用 Python#
from evalscope import run_task
from evalscope.config import TaskConfig
task_cfg = TaskConfig(
model='YOUR_MODEL',
api_url='OPENAI_API_COMPAT_URL',
api_key='EMPTY_TOKEN',
datasets=['swe_bench_verified_agentic'],
dataset_args={
'swe_bench_verified_agentic': {
# extra_params: {} # 使用默认额外参数
}
},
limit=10, # 正式评估时请删除此行
)
run_task(task_cfg=task_cfg)