腾讯云大模型训推平台TI-ONE 提交训练任务

相关产品

腾讯云大模型训推平台TI-ONE 产品页，含版本规格、价格与购买入口。

describe_postpaid_training_price(self) 查询每种配置的每小时的价格，价格单位：元 :rtype: tencentcloud.tione.v20211111.models.DescribeBillingSpecsResponsedescribe_system_reasoning_images(self) 获取平台内置的推理镜像 :return: 推理镜像信息 :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.DescribeInferTemplatesResponse` 数据格式如： { "FrameworkTemplates": [ { "Framework": "TENSORFLOW", "FrameworkVersion": "2.4", "Groups": [ "TENSORFLOW", "LIGHT" ], "InferTemplates": [ { "InferTemplateId": "tf2.4-py38-cpu", "InferTemplateImage": "ccr.ccs.tencentyun.com/qcloud-ti-platform/ti-cloud-infer-tensorflow-cpu:py38-tensorflow2.4-cpu-20211206" }, { "InferTemplateId": "tf2.4-py38-gpu", "InferTemplateImage": "ccr.ccs.tencentyun.com/qcloud-ti-platform/ti-cloud-infer-tensorflow-gpu:py38-tensorflow2.4-cu110-20211206" } ] } ], "RequestId": "3654e19b-c2ba-4953-b131-d66495723008" } def create_training_task(self, name, framework, cos_output, worker_resource, code_package_path, ps_resource=None, input_data_config=None, worker_start_cmd=None, ps_start_cmd=None, tags=None, tuning_parameters_dict={}, resource_group_id="", remark=None, log_enable=False, log_logset_id=None, log_topic_id=None, vpc_id=None, sub_net_id=None, restart_limit=0): """创建训练任务 :param name: 任务名称 :type name: str :param framework: 运行的框架环境 :type framework: :class:`tikit.models.FrameworkInfo` :param cos_output: 输出的cos信息 :type cos_output: str :param worker_resource: worker节点的配置 :type worker_resource: :class:`tikit.models.ResourceConfigInfo` :param code_package_path: 代码的cos信息 :type code_package_path: str :param ps_resource: ps节点的配置 :type ps_resource: :class:`tikit.models.ResourceConfigInfo` :param input_data_config: 输入的数据信息 :type input_data_config: list or :class:`tikit.models.TrainingDataConfig` :param worker_start_cmd: worker的启动命令 :type worker_start_cmd: str :param ps_start_cmd: ps节点的启动命令 :type ps_start_cmd: str :param tags: 标签 :type tags: list of :class:`tikit.tencentcloud.tione.v20211111.models.Tag` :param tuning_parameters_dict: 调参字典 :type tuning_parameters_dict: dict :param resource_group_id: 预付费的资源组id :type resource_group_id: str :param remark: 描述 :type remark: str :param log_enable: 日志开关 :type log_enable: bool :param log_logset_id: 日志集id :type log_logset_id: str :param log_topic_id: 日志的topic id :type log_topic_id: str :param vpc_id: vpc的id :type vpc_id: str :param sub_net_id: 子网id :type sub_net_id: str :param restart_limit: 当前任务最大重启次数，最高10次，超过后任务被标记为异常 :type restart_limit: int :return: :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.CreateTrainingTaskResponse` """def describe_training_frameworks(self): """查看训练框架 :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.DescribeTrainingFrameworksResponse` """def describe_training_tasks(self, filters=None, tag_filters=None, offset=0, limit=50, order="DESC", order_field="UpdateTime"): """获取训练任务列表 :param filters: 过滤器，eg：[{ "Name": "TaskStatus", "Values": ["Running"] }] :type filters: list of Filter :param tag_filters: 标签过滤器，eg：[{ "TagKey": "TagKeyA", "TagValue": ["TagValueA"] }] :type tag_filters: list of TagFilter :param offset: 偏移量，默认为0 :type offset: int :param limit: 返回数量，默认为50 :type limit: int :param order: 输出列表的排列顺序。取值范围：ASC：升序排列 DESC：降序排列 :type order: str :param order_field: 排序的依据字段，取值范围 "CreateTime" "UpdateTime" :type order_field: str :return: :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.DescribeTrainingTasksResponse` """def describe_training_task(self, task_id) 获取单个训练任务信息 :param task_id: 训练任务ID :type task_id: str :rtype: :class:`tencentcloud.tione.v20211111.models.DescribeTrainingTaskResponse`def describe_training_task_pods(self, task_id) 获取训练任务的pod列表 :param task_id: 训练任务ID :type task_id: str :rtype: :class:`tencentcloud.tione.v20211111.models.DescribeTrainingTaskPodsResponse`def describe_train_logs(self, pod_name, start_time=None, end_time=None, limit=None, order=None, context=None, filters=None): """查看训练任务的日志 :param pod_name: 查询哪个Pod的日志，支持通配符。查看某个训练任务的全部pod的日志可以填： "<task_id>-*"，如：train-51cd6bf7ec1000-* :type pod_name: str :param start_time: 日志查询开始时间。RFC3339格式的时间字符串，比如2021-12-16T13:20:24+08:00，默认值为当前时间的前一个小时 :type start_time: str :param end_time: 日志查询结束时间。RFC3339格式的时间字符串，比如2021-12-16T13:20:24+08:00，默认值为当前时间 :type end_time: str :param limit: 日志查询条数，默认值100，最大值100 :type limit: int :param order: 排序方向。(ASC | DESC) 默认值为DESC :type order: str :param context: 分页的游标 :type context: str :param filters: 过滤Filters :type filters: list of tikit.tencentcloud.tione.v20211111.models.Filter :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.DescribeLogsResponse` 返回的对象如果非空，就会有 next() 方法，能不断地获取下一页的日志（如果有多页的话），如下： now_time = datetime.datetime.now(datetime.timezone.utc) now_time_str = now_time.isoformat() result = client.describe_train_logs("train-51cd6bf7ec1000-37c5p5nlr01s-launcher", "2021-12-10T09:32:03.823509+00:00", now_time_str, limit=30) print(result) print(result.next()) print(result.next()) print(result.next()) """def push_training_metrics(self, timestamp, value_map, task_id=None, epoch=None, total_steps=None, step=None): """上报训练自定义指标（单条）。单个子账号每秒可以调用20次，请在你的训练代码中注意控制上报频率，避免超限报错。或者使用push_training_metrics_list :param timestamp: 时间戳 :type timestamp: int :param value_map: 指标映射。指标名称 -> 指标值 :type value_map: map: str -> float :param task_id: 任务ID。若为空，就当前取任务节点环境的 TI_TASK_ID 环境变量的值 :type task_id: str :param epoch: epoch值 :type epoch: int :param total_steps: 总步数 :type total_steps: int :param step: 第几步 :type step: int :return: :rtype: :class:`tikit.tencentcloud.tione.v20211111.models.PushTrainingMetricsResponse` client.push_training_metrics(int(time.time()), {"field1": 11, "field2": 12}, "task-id-00001", 3, 1000, 66) """def push_training_metrics_list(self, metric_list) 上报训练自定义指标（列表） :param metric_list: MetricData 数组。若任务ID为空，就当前取任务节点环境的 TI_TASK_ID 环境变量的值 :type metric_list: list of :class:`tencentcloud.tione.v20211111.models.MetricData` :return: :rtype: :class:`tencentcloud.tione.v20211111.models.PushTrainingMetricsResponse`def describe_training_metrics(self, task_id) 查询训练自定义指标 :param task_id: 任务ID :type task_id: str :rtype: :class:`tencentcloud.tione.v20211111.models.DescribeTrainingMetricsResponse`def stop_training_task(self, task_id) 停止某个训练任务 :param task_id: 训练任务ID :type task_id: str :rtype: :class:`tencentcloud.tione.v20211111.models.StopTrainingTaskResponse`def delete_training_task(self, task_id) 删除某个训练任务 :param task_id: 训练任务ID :type task_id: str :rtype: :class:`tencentcloud.tione.v20211111.models.DeleteTrainingTaskResponse`中间配置：tikit.models.FrameworkInfodef new_custom(training_mode, image_type, image_url, registry_region=None, registry_id=None, user_name=None, passwd=None): """自定义训练框架的配置通过describe_training_frameworks()查看列表 :param training_mode: 训练模式，如"DDP"，仅训练任务需要配置 :type training_mode: str :param image_type: 镜像类型，CCR为腾讯云容器镜像服务个人版，TCR为腾讯云容器镜像服务企业版，CUSTOM表示第三方自定义镜像 :type image_type: str :param image_url: 镜像地址必填 :type image_url: str :param registry_region: 腾讯云容器镜像服务的镜像仓库的地域 :type registry_region: str :param registry_id: 腾讯云容器镜像服务的镜像仓库ID :type registry_id: str :param user_name: 自定义镜像仓库的用户名 :type user_name: str :param passwd: 自定义镜像仓库的密码 :type passwd: str :return: :rtype: """def new_system_framework(framework_name, framework_environment, training_mode): """系统内置的训练框架 :param framework_name: 框架名称。通过describe_training_frameworks()查看列表 :type framework_name: str :param framework_environment: 框架环境。通过describe_training_frameworks()查看列表 :type framework_environment: str :param training_mode: 训练模式。通过describe_training_frameworks()查看列表 :type training_mode: str :return: :rtype: """—————————————————-tikit.models.ResourceConfigInfodef new_postpaid(instance_type, instance_num): """获取后付费模式下的资源配置 :param instance_type: 实例类型。通过 describe_postpaid_training_price() 查看实例列表 :type instance_type: str :param instance_num: 实例数量 :type instance_num: int :return: :rtype: """| def new_prepaid(cpu=0, memory=0, gpu=0, gpu_type=None, instance_num=1): """获取预付费模式下的资源配置，当资源组开启GPU按比例预设后，cpu和memory参数需要不设置或者设置为0，该两项的值将通过GPU卡数按照比例分配 :param cpu: CPU个数，单位是核 :type cpu: float :param memory: 内存大小，单位是GB :type memory: float :param gpu_type: gpu类型 :type gpu_type: str :param gpu: gpu个数 :type gpu: float :param instance_num: 实例数量 :type instance_num: int :return: :rtype: """—————————————————-tikit.models.TrainingDataConfigdef new_mount_cos(cos_str, target_path): """一个cos类型的训练数据 :param cos_str: cos存储，格式： <bucket>/<cos path>/ :type cos_str: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """def new_dataset_mount(dataset_id, target_path): """一个dataset类型的训练数据 :param dataset_id: 数据集ID :type dataset_id: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """def new_mount_cfs(cfs_id, source_path, target_path): """新建一个cfs类型的训练数据集配置 :param cfs_id: CFS的ID :type cfs_id: str :param source_path: CFS的路径 :type source_path: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """def new_mount_hdfs(hdfs_id, source_path, target_path): """新建一个hdfs类型的训练数据集配置 :param hdfs_id: EMR上HDFS的ID :type hdfs_id: str :param source_path: HDFS的路径 :type source_path: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """def new_mount_wedata_hdfs(wedata_id, source_path): """新建一个wedata hdfs类型的训练数据集配置 :param wedata_id: wedata数据源id :type wedata_id: int :param source_path: HDFS的路径 :type source_path: str :return: :rtype: """def new_dataset(id_target_dict): """ Deprecated ! 新建一个dataset类型的训练数据集配置 :param id_target_dict: 数据集信息。 dataset id -> 下载的目标路径 :type id_target_dict: dict :return: :rtype: """def new_cos_data(cos_str_target_dict): """Deprecated ! 新建一个cos类型的训练数据集配置 :param cos_str_target_dict: 数据集信息。 <bucket>/<cos path>/ -> 下载的目标路径 :type cos_str_target_dict: dict :return: :rtype: """def new_mount_goosefs(goosefs_id, source_path, namespace, target_path): """新建一个goosefs类型的训练数据集配置 :param goosefs_id: goosefs实例id :type goosefs_id: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """ def new_mount_goosefsx(goosefsx_id, goosefsx_path, target_path): """新建一个goosefsx类型的存储配置 :param goosefsx_id: goosefsx实例id :type goosefsx_id: str :param goosefsx_path: goosefsx路径 :type goosefsx_path: str :param target_path: 目标挂载路径 :type target_path: str :return: :rtype: """

腾讯云大模型训推平台TI-ONE 提交训练任务

相关内容：