1. 接口描述
接口请求域名: dlc.tencentcloudapi.com 。
断点续训(克隆实例)
默认接口请求频率限制:20次/秒。
2. 输入参数
以下请求参数列表仅列出了接口请求参数和部分公共参数,完整公共参数列表见 公共请求参数。
| 参数名称 | 必选 | 类型 | 描述 |
|---|---|---|---|
| Action | 是 | String | 公共参数,本接口取值:SubmitTrainingJob。 |
| Version | 是 | String | 公共参数,本接口取值:2021-01-25。 |
| Region | 是 | String | 公共参数,详见产品支持的 地域列表,本接口仅支持其中的: ap-beijing, ap-chongqing, ap-guangzhou, ap-shanghai, ap-singapore 。 |
| SpecName | 否 | String | 训练作业配置名称(≤255 字符) 示例值:cus0724-2 |
| Description | 否 | String | 描述(≤1024 字符) 示例值:cus04 test |
| Entrypoint | 否 | String | 启动命令 示例值:bash -c 'cd sft_demo && python -m sft_demo.train_local --storage-path /shared' |
| Image | 否 | String | 镜像地址 示例值:ccr.ccs.tencentyun.com/emr-image/tcray:2.55.1-py311-cpu-torch-lab |
| ImagePullType | 否 | String | 镜像拉取类型(BuiltIn: 内置, Custom: 自定义-TCR, CustomCcr: 自定义-CCR) 示例值:BuiltIn |
| ImagePullPolicy | 否 | String | 镜像拉取策略(Always / IfNotPresent / Never) 示例值:IfNotPresent |
| CodePackageUrl | 否 | String | 代码包 COS URL 示例值:https://common-job-packages-251233710.cos.ap-guangzhou.myqcloud.com/job-packages/260200065/20260724134025/b5555d51-563c-4519-b5b5-c57c53e6b3eb.zip |
| RuntimeEnv | 否 | String | Ray runtime_env 配置 JSON(含 pip 依赖、env_vars 等,结构参见 2.1) 示例值:{ "env_vars": { "LEARNING_RATE": "1e-4", "BATCH_SIZE": "32" }, "pip": [] } |
| ResourceConfigId | 否 | String | 资源配置模板 ID(可选) 示例值:0453d291-7b46-4b04-a989-f659d419b8a5 |
| ResourceConfig | 否 | String | 资源配置 JSON 示例值:{"Head":{"Name":"default-head","PodCpu":2,"PodMem":8,"GpuType":"","GpuNum":0,"Envs":[],"PodNum":1,"HighAvailability":false},"Worker":[{"Name":"default-worker","PodCpu":2,"PodMem":8,"GpuType":"","GpuNum":0,"Envs":[],"MinPodNum":1,"MaxPodNum":1}]} |
| ResourcePartitionId | 否 | String | 资源分区 ID 示例值:dlc-p-wdtiljwu |
| Queue | 否 | String | 队列名称 示例值:default |
| Catalog | 否 | String | 存储卷挂载配置 JSON(含 Source 字段标记用途) 示例值:{"COSVolumes":[{"VolumeSubPath":"/cus0724-2","MountPath":"/shared","Source":"checkpoint","Bucket":"common-job-packages-251233710"}]} |
| AdvancedOptions | 否 | String | 高级参数 JSON(不传则不更新) 示例值:{"spec.suspend":"false"} |
| BaseModelUid | 否 | String | 基础模型Uid 示例值:m-test1-6a1e7ed1-33de |
| Mode | 否 | String | 算法模式:sft / dpo / cpt / grpo(仅 POST_TRAINING 必填,CUSTOM_CODE / LAB 禁止传入) 示例值:sft |
| Datasets.N | 否 | Array of DatasetMount | 数据集挂载列表(元素含 DatasetId 或 Catalog 二选一 + DatasetName + Eval 属性) |
| Checkpoint | 否 | CheckpointConfig | Checkpoint 产出配置(POST_TRAINING 必填;CUSTOM_CODE / LAB 可选) |
| ResumeTraining | 否 | Boolean | 是否启用断点续训 示例值:false |
| TuningParams | 否 | TrainingTuningParams | 调优参数(高级参数,仅 POST_TRAINING 使用;CUSTOM_CODE / LAB 禁止传入) |
| Priority | 否 | Integer | 作业优先级(1-9,数字越大优先级越高) 示例值:5 |
| Kind | 否 | String | 提交来源标签:LAB / CUSTOM_CODE(可选,用于溯源,不影响处理逻辑) 示例值:CUSTOM_CODE |
| MlFlowConfig | 否 | MlFlowConfig | MlFlow 实验追踪配置(可选,不传则不启用 MlFlow) |
| Tags.N | 否 | Array of Tag | 标签列表(TagKey-TagValue),用于将任务与腾讯云标签系统中的标签绑定 |
| OutputModelName | 否 | String | 产出模型名称(用于后续模型注册,当前仅保存) 示例值:model-test |
3. 输出参数
| 参数名称 | 类型 | 描述 |
|---|---|---|
| Spec | TrainingJobSpec | 训练作业配置详情 |
| RequestId | String | 唯一请求 ID,由服务端生成,每次请求都会返回(若请求因其他原因未能抵达服务端,则该次请求不会获得 RequestId)。定位问题时需要提供该次请求的 RequestId。 |
4. 示例
示例1 训练任务统一提交入口
输入示例
POST / HTTP/1.1
Host: dlc.tencentcloudapi.com
Content-Type: application/json
X-TC-Action: SubmitTrainingJob
<公共请求参数>
{
"SpecName": "cus0724-2",
"Description": "",
"Entrypoint": "bash -c 'cd sft_demo && python -m sft_demo.train_local --storage-path /shared'",
"Image": "ccr.ccs.tencentyun.com/emr-image/tcray:2.55.1-py311-cpu-torch-lab",
"ImagePullType": "BuiltIn",
"ImagePullPolicy": "IfNotPresent",
"CodePackageUrl": "https://common-job-packages-251233710.cos.ap-guangzhou.myqcloud.com/job-packages/260200065/20260724134025/b5555d51-563c-4519-b5b5-c57c53e6b3eb.zip",
"ResourceConfig": "{\"Head\":{\"Name\":\"default-head\",\"PodCpu\":2,\"PodMem\":8,\"GpuType\":\"\",\"GpuNum\":0,\"Envs\":[],\"Labels\":[],\"PodNum\":1,\"HighAvailability\":false},\"Worker\":[{\"Name\":\"default-worker\",\"PodCpu\":2,\"PodMem\":8,\"GpuType\":\"\",\"GpuNum\":0,\"Envs\":[],\"Labels\":[],\"MinPodNum\":1,\"MaxPodNum\":1}]}",
"ResourcePartitionId": "dlc-p-wdtiljwu",
"Queue": "default",
"Catalog": "{\"COSVolumes\":[{\"VolumeSubPath\":\"/cus0724-2\",\"MountPath\":\"/shared\",\"Source\":\"checkpoint\",\"Bucket\":\"common-job-packages-251233710\"}]}",
"Priority": 5,
"Kind": "CUSTOM_CODE",
"OutputModelName": "model-test"
}
输出示例
{
"Response": {
"Spec": {
"Catalog": "{\"COSVolumes\":[{\"VolumeSubPath\":\"/cus0724-2\",\"MountPath\":\"/shared\",\"Source\":\"checkpoint\",\"Bucket\":\"common-job-packages-251233710\"}]}",
"CheckpointMountInfo": {
"Bucket": "common-job-packages-251233710",
"MountPath": "/shared",
"PlatformManaged": true,
"Region": "ap-guangzhou",
"StorageType": "COS",
"VolumeSubPath": "/cus0724-2"
},
"CodePackageUrl": "https://common-job-packages-251233710.cos.ap-guangzhou.myqcloud.com/job-packages/260200065/20260724134025/b5555d51-563c-4519-b5b5-c57c53e6b3eb.zip",
"CreateTime": 1784884416974,
"Creator": "700002655693",
"Description": "",
"Entrypoint": "bash -c 'cd sft_demo && python -m sft_demo.train_local --storage-path /shared'",
"HasRunningInstances": false,
"Image": "ccr.ccs.tencentyun.com/emr-image/tcray:2.55.1-py311-cpu-torch-lab",
"ImagePullPolicy": "IfNotPresent",
"ImagePullType": "BuiltIn",
"InstanceCount": 0,
"Kind": "CUSTOM_CODE",
"OutputModelName": "model-test",
"Priority": 5,
"Queue": "default",
"ResourceConfig": "{\"Head\":{\"Name\":\"default-head\",\"PodCpu\":2,\"PodMem\":8,\"GpuType\":\"\",\"GpuNum\":0,\"Envs\":[],\"Labels\":[],\"PodNum\":1,\"HighAvailability\":false},\"Worker\":[{\"Name\":\"default-worker\",\"PodCpu\":2,\"PodMem\":8,\"GpuType\":\"\",\"GpuNum\":0,\"Envs\":[],\"Labels\":[],\"MinPodNum\":1,\"MaxPodNum\":1}]}",
"ResourcePartitionId": "dlc-p-wdtiljwu",
"ResourcePartitionName": "leion",
"SpecId": "raytrain-spec-tio9mo-n1jr",
"SpecName": "cus0724-2",
"UpdateTime": 1784884416974
},
"RequestId": "f3f6421b-da5b-4ab0-8860-15202a940ed6"
}
}
5. 开发者资源
腾讯云 API 平台
腾讯云 API 平台 是综合 API 文档、错误码、API Explorer 及 SDK 等资源的统一查询平台,方便您从同一入口查询及使用腾讯云提供的所有 API 服务。
API Inspector
用户可通过 API Inspector 查看控制台每一步操作关联的 API 调用情况,并自动生成各语言版本的 API 代码,也可前往 API Explorer 进行在线调试。
SDK
云 API 3.0 提供了配套的开发工具集(SDK),支持多种编程语言,能更方便的调用 API。
- Tencent Cloud SDK 3.0 for Python: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for Java: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for PHP: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for Go: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for Node.js: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for .NET: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for C++: CNB, GitHub, Gitee
- Tencent Cloud SDK 3.0 for Ruby: CNB, GitHub, Gitee
命令行工具
6. 错误码
该接口暂无业务逻辑相关的错误码,其他错误码详见 公共错误码。