ServerlessJobConfig

The base model Amazon Resource Name (ARN) in SageMaker Public Hub. SageMaker always selects the latest version of the provided model.

Type: String

Length Constraints: Minimum length of 1. Maximum length of 2048.

Pattern: (arn:[a-z0-9-\.]{1,63}:sagemaker:\w+(?:-\w+)+:(\d{12}|aws):hub-content\/)[a-zA-Z0-9](-*[a-zA-Z0-9]){0,62}\/Model\/[a-zA-Z0-9](-*[a-zA-Z0-9]){0,63}(\/\d{1,4}.\d{1,4}.\d{1,4})?

Required: Yes

The serverless training job type.

Type: String

Valid Values: FineTuning | Evaluation

Required: Yes

Specifies agreement to the model end-user license agreement (EULA). The AcceptEula value must be explicitly defined as True in order to accept the EULA that this model requires. You are responsible for reviewing and complying with any applicable license terms and making sure they are acceptable for your use case before downloading or using a model. For more information, see End-user license agreements section for more details on accepting the EULA.

Type: Boolean

Required: No

The model customization technique.

Type: String

Valid Values: SFT | DPO | RLVR | RLAIF

Required: No

The evaluation job type. Required when serverless job type is Evaluation.

Type: String

Valid Values: LLMAJEvaluation | CustomScorerEvaluation | BenchmarkEvaluation

Required: No

The evaluator Amazon Resource Name (ARN) used as reward function or reward prompt.

Type: String

Pattern: .*

Required: No

The parameter-efficient fine-tuning configuration.

Type: String

Valid Values: LORA

Required: No

The maximum sequence length, in tokens, that the customization job supports. SageMaker uses this value to select a training configuration for the base model that you specify. The parameter supports the following values:

If you don't specify a value, SageMaker selects a training configuration based on the other values that you specify. The selection is not restricted to a particular sequence length.

Type: String

Length Constraints: Minimum length of 1. Maximum length of 50.

Pattern: \d+K

Required: No