导航
AgentPilot OpenAPI 使用指南
最近更新时间：2025.10.16 15:57:03首次发布时间：2025.10.16 15:57:03
文档反馈
问问助手环境变量配置
- 配置 API_KEY 和 API_URL ：请正确配置 API_KEY 和 API_URL，确保 API 请求的正常发送与接收。
- Workspace ID 获取与配置 ：当前版本暂时支持从环境变量配置 workspace id ，接口暂不支持每次调用单独指定。获取方式为在 PromptPilot 页面切换到某个个人或团队空间后，从顶部 URL 截取 "workspaceId=" 之后的内容，即为当前 workspace_id。后续预计会更新 SDK，支持通过参数传入 workspace id 。
export AGENTPILOT_API_URL=https://prompt-pilot.cn-beijing.volces.com export AGENTPILOT_API_KEY=<YOUR API KEY FROM PROMPTPILOT CONSOLE> export AGENTPILOT_WORKSPACE_ID=<YOUR WORKSPACE ID FROM PROMPTPILOT CONSOLE>
Task 和 Prompt 管理
创建任务
创建任务后，可登录 PromptPilot Console 页面，在任务管理中查看最新创建的任务。
创建文本理解任务
所有 model_name 查看https://www.volcengine.com/docs/82379/1330310 ，成功后将看到创建后的任务信息和 Prompt 版本信息。
# Create a new task of text understanding curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=CreateTask" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_category": "DEFAULT", "task_name": "test_text_task", "messages": [ { "role": "user", "content": "这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}" } ], "model_name": "doubao-seed-1.6-250615", "criteria": "这是一个文本理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确" }'
如果成功，你将看到创建后的任务信息和Prompt版本信息
{"Result":{"success":true,"prompt_version":{"prompt_version_id":"prv-20250905104255-nafnT","task_id":"ta-20250905104255-r8u5s","version":"v1","messages":[{"content":[{"text":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}","type":"text"}],"role":"user"}],"variables_name":["variable_1","variable_2"],"variables_types":null,"model_name":"doubao-seed-1.6-250615","temperature":1.0,"top_p":0.7,"eval_dimension":"这是一个文本理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确","created_at":"2025-08-28T14:58:56.938785","updated_at":"2025-08-28T14:58:56.938811"}},"Error":null}
创建视觉理解任务
# Create a new task of visual understanding curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=CreateTask" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_category": "MULTIMODAL", "task_name": "test_visual_task", "messages": [ { "role": "user", "content": "根据文字 {{text_var}} 和配图 {{image_var}}，写一个小故事。" } ], "variables_types": { "text_var": "text", "image_var": "image_url" }, "model_name": "doubao-seed-1.6-250615", "criteria": "这是一个视觉理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确" }'
如果成功，你将看到创建后的任务信息和Prompt版本信息
{"Result":{"success":true,"prompt_version":{"prompt_version_id":"prv-20250905104357-mV3yD","task_id":"ta-20250905104357-uDfgL","version":"v1","messages":[{"role":"user","content":[{"type":"text","text":"根据文字 {{text_var}} 和配图 "},{"type":"image_url","image_url":{"url":"{{image_var}}"}},{"type":"text","text":"，写一个小故事。"}]}],"variables_name":["text_var","image_var"],"variables_types":{"text_var":"text","image_var":"image"},"model_name":"doubao-seed-1.6-250615","temperature":1.0,"top_p":0.7,"eval_dimension":"这是一个视觉理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确","created_at":"2025-08-28T15:01:32.633732","updated_at":"2025-08-28T15:01:32.633755"}},"Error":null}
创建多轮对话任务
# Create a new task of dialog curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=CreateTask" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_category": "DIALOG", "task_name": "test_dialog_task", "task_name": "test_dialog_task", "messages": [ { "role": "system", "content": "你是一个MBTI人格分析大师，负责根据用户提供的个人信息分析用户的MBTI人格。" } ], "model_name": "doubao-seed-1.6-250615", "criteria": "这是一个多轮对话任务的评分标准，5分制，1分表示完全错误，5分表示完全正确" }'
如果成功，你将看到创建后的任务信息和Prompt版本信息
{"Result":{"success":true,"prompt_version":{"prompt_version_id":"prv-20250905104417-bsqcQ","task_id":"ta-20250905104417-aaQfG","version":"v1","messages":[{"content":[{"text":"你是一个MBTI人格分析大师，负责根据用户提供的个人信息分析用户的MBTI人格。","type":"text"}],"role":"system"}],"variables_name":[],"variables_types":null,"model_name":"doubao-seed-1.6-250615","temperature":1.0,"top_p":0.7,"eval_dimension":"这是一个多轮对话任务的评分标准，5分制，1分表示完全错误，5分表示完全正确","created_at":"2025-08-28T15:01:28.277163","updated_at":"2025-08-28T15:01:28.277196"}},"Error":null}
查看某一个 version 的信息
将示例代码中的 task_id 和 version 分别替换为实际的 task_id 和 version 后运行，成功后将看到该 task_id 下指定 version 的 Prompt 版本信息。
# Set task_id and version obtained from above commands or from PromptPilot Console UI export task_id="ta-20250905101023-qEeWK" export version="v1" # Get a prompt version curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=GetPromptTemplate" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_id": "'$task_id'", "version": "'$version'" }'
{"Result":{"success":true,"prompt_version":{"prompt_version_id":"prv-20250905104255-nafnT","task_id":"ta-20250905104255-r8u5s","version":"v1","messages":[{"content":[{"text":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}","type":"text"}],"role":"user"}],"variables_name":["variable_1","variable_2"],"variables_types":null,"model_name":"doubao-seed-1.6-250615","temperature":1.0,"top_p":0.7,"eval_dimension":"这是一个文本理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确","created_at":"2025-09-05T10:42:55","updated_at":"2025-09-05T10:42:55"}},"Error":null}
列出一个 task 下的所有版本信息
将示例代码中的 task_id 替换为实际的 task_id 后运行，成功后将看到该 task_id 下的所有 Prompt 版本信息。
# Set task_id obtained from above commands or from PromptPilot Console UI export task_id="ta-20250905101023-qEeWK" # List all prompt versions of a task curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=ListPromptTemplates" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_id": "'$task_id'" }'
{"Result":{"success":true,"prompt_versions":[{"prompt_version_id":"prv-20250905104255-nafnT","task_id":"ta-20250905104255-r8u5s","version":"v1","messages":[{"content":[{"text":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}","type":"text"}],"role":"user"}],"variables_name":["variable_1","variable_2"],"variables_types":null,"model_name":"doubao-seed-1.6-250615","temperature":1.0,"top_p":0.7,"eval_dimension":"这是一个文本理解任务的评分标准，5分制，1分表示完全错误，5分表示完全正确","created_at":"2025-08-28T15:01:25.470669","updated_at":"2025-08-28T15:01:25.470694"}]},"Error":null}
流式生成 Prompt
文本理解
# Test text understanding tasks curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=GeneratePromptStream" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_type": "DEFAULT", "rule": "从提供的文档中提取总结要点，要点数量不超过10个。", "model_name": "doubao-seed-1.6-250615", "temperature": 1.0, "top_p": 0.7 }'
如果成功，你将在Terminal看到流式生成的Prompt结果。
视觉理解
# Test visual understanding tasks curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=GeneratePromptStream" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_type": "MULTIMODAL", "rule": "依据提供给的题目图片，对其中的题目展开解答。", "model_name": "doubao-seed-1.6-250615", "temperature": 1.0, "top_p": 0.7 }'
如果成功，你将在Terminal看到流式生成的Prompt结果。
多轮对话
# Test dialog tasks curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=GeneratePromptStream" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_type": "DIALOG", "rule": "你是一个MBTI人格分析大师，负责根据用户提供的个人信息分析用户的MBTI人格。", "model_name": "doubao-seed-1.6-250615", "temperature": 1.0, "top_p": 0.7 }'
如果成功，你将在Terminal看到流式生成的Prompt结果。
回流数据和反馈
数据回流范围
可以将变量、图片、对话、模型回答、参照回答、评分及评分原因等各类事件回流到平台。在 PromptPilot Console 页面对应的 task_id 和 version 的批量评测页可查看回流的数据。
上传方式
- 通过完整 curl 命令上传 ：将示例代码中的 task_id 替换为实际的 task_id 后运行，成功后将看到返回后的状态，包括每条数据的上传状态。
# Assume you have already known a task_id and one of its version export task_id="ta-20250904134351-AUtEk" export version="v1" # Upload a test dataset with 2 data records curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=TrackingEvent" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "tracking_events": [ { "run_id": "'$(uuidgen)'", "run_type": "llm", "event_name": "upload a record", "task_id": "'$task_id'", "prompt_version": "'$version'", "variables": { "variable_1": "value_1", "variable_2": "value_2" }, "output_message": { "role": "assistant", "content": "这是一个文本理解任务的model answer for value 1 and value 2" }, "reference": { "role": "assitant", "content": "这是一个文本理解任务的reference answer for value 1 and value 2" }, "feedback": { "human_score": 5, "human_analysis": "该条数据的人工评分原因", "human_confidence": 1.0 } }, { "run_id": "'$(uuidgen)'", "run_type": "llm", "event_name": "upload a record", "task_id": "'$task_id'", "prompt_version": "'$version'", "variables": { "variable_1": "value_3", "variable_2": "value_4" }, "output_message": { "role": "assistant", "content": "这是一个文本理解任务的model answer for value 3 and value 4" }, "reference": { "role": "assitant", "content": "这是一个文本理解任务的reference answer for value 3 and value 4" }, "feedback": { "human_score": 1, "human_analysis": "该条数据的人工评分原因", "human_confidence": 1.0 } } ] }'
{"Result":{"success":true,"tracking_events":{"E5C683EB-B0F7-4B2F-BAFC-7AF77AA90666":{"task_id":"ta-20250905104255-r8u5s","prompt_version":"v1","run_id":"E5C683EB-B0F7-4B2F-BAFC-7AF77AA90666","session_id":null,"status":{"success":true,"error":null}},"D005BE7E-7E15-46FF-85D9-DE35A6FB4F4F":{"task_id":"ta-20250905104255-r8u5s","prompt_version":"v1","run_id":"D005BE7E-7E15-46FF-85D9-DE35A6FB4F4F","session_id":null,"status":{"success":true,"error":null}}}},"Error":null}
此时，登陆PromptPilot Console页面的批量评测，你将看到回流的数据
- 通过 JSON 文件上传数据集 ：
- 下载示例文件AgentPilot Open API - 发布版到某一路径下。
- 在该文件所在路径运行指定命令，或修改为该文件所在绝对路径。注意需将示例 JSON 文件中的 task_id 替换为自己的 task_id。
# Alternatively, you can upload the test dataset in a JSON file. # You need to set the related fields before submit a request, which may include request_id, task_id, version, run_id, etc. curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=TrackingEvent" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '@./test_dataset.json'
Prompt优化和报告获取
提交优化任务
对某个task_id任务下的某个version提交优化job，将task_id和version分别替换为你的task_id和version
# Assume you have already known a task_id and one of its version export task_id="ta-20250904134351-AUtEk" export version="v1" # Submit an optimization job for the task_id of the version. curl --location "$AGENTPILOT_API_URL/agent-pilot-optimize?Version=2024-01-01&Action=OptimizeServiceStartOptimize" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_id": "'$task_id'", "version": "'$version'" }'
如果成功，你将得到optimize_job_id，通过该optimize_job_id，可以帮助你
- 查询优化任务的进展，
- 监控中间结果，
- 以及获取优化报告
{"Result":{"TaskId":"ta-20250905104255-r8u5s","Version":"v1","OptimizeJobId":"20250905105402_XGtu9_OptimizePrompt"},"Error":null}
查看任务进展
将下面optimize_job_id替换为你的optimize_job_id后运行
export optimize_job_id="20250904152330_zUuSs_OptimizePrompt" # Check progress during the optimization (State=3) or after the optimization (State=7) curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=OptimizeServiceGetOptimizeProgress" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "optimize_job_id": "'$optimize_job_id'" }'
State | 解释 |
|---|---|
1 | 优化任务已创建 |
2 | 优化任务运行中 |
3 | 优化任务已完成且优化成功 |
4 | 优化任务已完成且优化失败 |
比如运行中 (State=2) 如下图
{"Result":{"JobInfo":{"JobId":"20250904211525_EvMYQ_OptimizePrompt","Version":"v1","State":2,"OptimizedVersion":null,"CreatedTime":"2025-09-05T13:15:25","UpdatedTime":"2025-09-05T13:15:35"},"Progress":{"ProgressPercent":0.0,"TotalCnt":2,"BetterCnt":0,"WorseCnt":0,"UnchangedCnt":2,"InitFullscoreCnt":1,"FullscoreCntList":[1],"InitAverageScore":3.0,"AverageScoreList":[3.0],"OptimizeTokenConsumption":0,"OptimalPrompt":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}"}},"Error":null}
比如优化成功 (State=3) 如下图
{"Result":{"JobInfo":{"JobId":"20250904211525_EvMYQ_OptimizePrompt","Version":"v1","State":3,"OptimizedVersion":"v2","CreatedTime":"2025-09-05T13:15:25","UpdatedTime":"2025-09-05T13:16:02"},"Progress":{"ProgressPercent":1.0,"TotalCnt":2,"BetterCnt":0,"WorseCnt":0,"UnchangedCnt":2,"InitFullscoreCnt":1,"FullscoreCntList":[1,1,1],"InitAverageScore":3.0,"AverageScoreList":[3.0,3.0,3.0],"OptimizeTokenConsumption":0,"OptimalPrompt":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}"}},"Error":null}
说明
你也可以登陆PromptPilot Console页面查看实时进展监控曲线
获取优化报告
分别将task_id，base_version, ref_version替换为你的优化任务的task_id, base_version，ref_version，你将得到你刚才提交的从base_version优化到ref_version的优化报告。
# Set the base and ref versions when the optimization progress is done (State=7). # The base version is "Version", and the "OptimizedVersion". export base_version="v1" # Version in previous command export ref_version="v2" # OptimizedVersion in previous command # Then, you can get the report. curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=OptimizeServiceGetReport" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "task_id": "'$task_id'", "base_version": "'$base_version'", "ref_version": "'$ref_version'" }'
如果运行成功，你将看到
{"Result":{"base":{"records":[{"record_id":"dr-20250904212324-ymasZ","variables":[{"name":"variable_1","value":"value_1","type":"text"},{"name":"variable_2","value":"value_2","type":"text"}],"model_answer":"这是一个文本理解任务的model answer for value 1 and value 2","ref_answer":"这是一个文本理解任务的reference answer for value 1 and value 2","score":5,"analysis":"该条数据的人工评分原因","confidence":null},{"record_id":"dr-20250904212326-LnpPS","variables":[{"name":"variable_1","value":"value_3","type":"text"},{"name":"variable_2","value":"value_4","type":"text"}],"model_answer":"这是一个文本理解任务的model answer for value 3 and value 4","ref_answer":"这是一个文本理解任务的reference answer for value 3 and value 4","score":1,"analysis":"该条数据的人工评分原因","confidence":null}],"prompt":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}","metric":"","avg_score":3.0},"ref":{"records":[{"record_id":"dr-20250904212605-Stwkk","variables":[{"name":"variable_1","value":"value_1","type":"text"},{"name":"variable_2","value":"value_2","type":"text"}],"model_answer":"这是一个文本理解任务的model answer for value 1 and value 2","ref_answer":"这是一个文本理解任务的reference answer for value 1 and value 2","score":5,"analysis":"该条数据的人工评分原因","confidence":1.0},{"record_id":"dr-20250904212605-49zY5","variables":[{"name":"variable_1","value":"value_3","type":"text"},{"name":"variable_2","value":"value_4","type":"text"}],"model_answer":"这是一个文本理解任务的model answer for value 3 and value 4","ref_answer":"这是一个文本理解任务的reference answer for value 3 and value 4","score":1,"analysis":"该条数据的人工评分原因","confidence":1.0}],"prompt":"这是一个文本理解任务的prompt，它有{{variable_1}}和{{variable_2}}","metric":"- 若回答完全未能结合输入的{{variable_1}}和{{variable_2}}，或者存在重大错误、误导性信息，得1分。\n- 若回答有错误、遗漏了输入信息相关的重要内容或表达不清晰，得2分。\n- 若回答基本结合了输入信息，内容正确但不够全面，细节或深度有所欠缺，得3分。\n- 若回答正确且较为完整地结合了输入信息，表达清晰，但还有可提升的空间，得4分。\n- 若回答准确、全面地结合了输入信息，表达清晰流畅，完全满足甚至超出预期要求，得5分。","avg_score":3.0}},"Error":null}
评估服务
提交一个评估请求
文本理解
curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=EvalServiceInputResponseEvaluate" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "eval_data_example": { "example_id": "example_001", "messages": [ { "role": "user", "content": "What is the capital of France?" } ], "response": [ { "role": "assistant", "content": "The capital of France is Paris." } ], "reference": [ { "role": "assistant", "content": "Paris is the capital city of France." } ], "score": null, "analysis": null, "confidence": null }, "metric_prompt": { "criteria": "Check if the response correctly identifies Paris as the capital of France." } }'
如果成功，你将看到如下格式的输出
{"Result":{"success":true,"evaled_data_example":{"example_id":"example_001","messages":[{"role":"user","content":"What is the capital of France?"}],"response":[{"role":"assistant","content":"The capital of France is Paris."}],"reference":[{"role":"assistant","content":"Paris is the capital city of France."}],"score":5,"analysis":"模型回答明确指出法国的首都是巴黎，与理想回答意思一致，正确完成了评估规则的要求，因此给予5分。","confidence":1.0}},"Error":null}
视觉理解
# Test the evaluation for a multimodal example curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=EvalServiceInputResponseEvaluate" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "eval_data_example": { "example_id": "example_001", "messages": [ { "role": "user", "content": [ {"text": "How many people are in the image?", "type": "text"}, {"image_url": {"url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lm_sth/ljhwZthlaukjlkulzlp/autope/examples/3.jpeg"}, "type": "image_url"} ] } ], "response": [ { "role": "assistant", "content": "There are 3 people in the image." } ], "reference": [ { "role": "assistant", "content": "There are 5 people in the image." } ], "score": null, "analysis": null, "confidence": null }, "metric_prompt": { "criteria": "Check if the response correctly counts the number of people in the image." } }'
如果成功，你将看到如下格式的输出
{"Result":{"success":true,"evaled_data_example":{"example_id":"example_001","messages":[{"role":"user","content":[{"type":"text","text":"How many people are in the image?"},{"type":"image_url","image_url":{"url":"https://lf3-static.bytednsdoc.com/obj/eden-cn/lm_sth/ljhwZthlaukjlkulzlp/autope/examples/3.jpeg"}}]}],"response":[{"role":"assistant","content":"There are 3 people in the image."}],"reference":[{"role":"assistant","content":"There are 5 people in the image."}],"score":1,"analysis":"根据评估规则，模型回答需要正确计算图片中的人数。理想回答是5人，而模型回答是3人，明显错误，未达到评估要求。","confidence":1.0}},"Error":null}
生成一个评估标准
文本理解
# Test the criteria generation for a set of examples curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=EvalServiceCriteriaGeneration" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "eval_data_examples": [ { "example_id": "example_001", "messages": [ { "role": "user", "content": "What is the weather like today?" } ], "response": [ { "role": "assistant", "content": "The weather is pleasant and warm." } ], "reference": [ { "role": "assistant", "content": "It is sunny with a high of 75 degrees Fahrenheit." } ], "score": 4, "analysis": "The response accurately describes the weather but could be more detailed.", "confidence": 0.85 }, { "example_id": "example_002", "messages": [ { "role": "user", "content": "Will it rain tomorrow?" } ], "response": [ { "role": "assistant", "content": "There is a chance of rain tomorrow." } ], "reference": [ { "role": "assistant", "content": "According to the forecast, there is a 60% chance of rain tomorrow with possible thunderstorms in the afternoon." } ], "score": 2, "analysis": "The response is too vague and lacks the specific details provided in the reference answer.", "confidence": 0.9 } ] }'
如果成功，你将看到如下格式的输出
{"Result":{"success":true,"generated_criteria":"\n回答与问题无关或完全错误，评为 1 分\n回答过于模糊，缺乏关键细节，评为 2 分\n能准确回答问题，包含一定关键信息，评为 3 分\n回答准确，但详细程度有提升空间，评为 4 分\n回答准确且详细全面，提供丰富相关信息，评为 5 分\n"},"Error":null}
视觉理解
curl --location "$AGENTPILOT_API_URL/agent-pilot?Version=2024-01-01&Action=EvalServiceCriteriaGeneration" \ --header "Authorization: Bearer $AGENTPILOT_API_KEY" \ --header "Content-Type: application/json" \ --data '{ "request_id": "'$(uuidgen)'", "workspace_id": "'$AGENTPILOT_WORKSPACE_ID'", "eval_data_examples": [ { "example_id": "example_001", "messages": [ { "role": "user", "content": [ {"text": "How many people are in the image?", "type": "text"}, {"image_url": {"url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lm_sth/ljhwZthlaukjlkulzlp/autope/examples/3.jpeg"}, "type": "image_url"} ] } ], "response": [ { "role": "assistant", "content": "There are 3 people in the image." } ], "reference": [ { "role": "assistant", "content": "There are 5 people in the image." } ], "score": 1, "analysis": "The response incorrectly counts the number of people in the image.", "confidence": 1.0 }, { "example_id": "example_002", "messages": [ { "role": "user", "content": [ {"text": "How many people are in the image?", "type": "text"}, {"image_url": {"url": "https://ark-auto-2100000825-cn-beijing-default.tos-cn-beijing.volces.com/experience_video_demo/experience_videofU0eQ1xX.png"}, "type": "image_url"} ] } ], "response": [ { "role": "assistant", "content": "There are 1 people in the image." } ], "reference": [ { "role": "assistant", "content": "There are 2 people in the image." } ], "score": 1, "analysis": "The response incorrectly counts the number of people in the image.", "confidence": 1.0 } ] }'
如果成功，你将看到如下格式的输出
{"Result":{"success":true,"generated_criteria":"\n模型回答的人数与图片实际人数完全一致，得5分。\n模型回答的人数与图片实际人数相差1人，得3分。\n模型回答的人数与图片实际人数相差2人及以上，得1分。\n"},"Error":null}