sagemaker.train.rft.models

Contract models for the rollout server API.

These models define the enforced contract between the platform trainer and customer rollout servers.

Customer server requirements:

POST /rollout - Accept RolloutRequest GET /health - Return {“status”: “healthy”} when ready

Classes

InferenceParams(*[, temperature, ...])	Inference parameters for rollout sampling.
RolloutMetadata(*, job_arn, trajectory_id, ...)	Metadata sent by the trainer with each rollout request.
RolloutRequest(*, instance, metadata[, ...])	Request format sent by the trainer to your /rollout endpoint.

class sagemaker.train.rft.models.InferenceParams(*, temperature: float | None = None, max_tokens: int | None = None, top_p: float | None = None)

Bases: BaseModel

Inference parameters for rollout sampling.

All fields are optional - if not provided, model defaults are used.

max_tokens: int | None

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

temperature: float | None

top_p: float | None

class sagemaker.train.rft.models.RolloutMetadata(*, job_arn: str, trajectory_id: str, endpoint: str, region: str = 'us-west-2')

Bases: BaseModel

Metadata sent by the trainer with each rollout request.

Pass this entire object (or its dict form) to RolloutFeedbackClient and make_inference_headers.

endpoint: str

job_arn: str

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

region: str

trajectory_id: str

class sagemaker.train.rft.models.RolloutRequest(*, instance: Dict[str, Any], metadata: RolloutMetadata, inference_params: InferenceParams | None = None, model_name: str | None = None, model_endpoint: str | None = None)

Bases: BaseModel

Request format sent by the trainer to your /rollout endpoint.

This is the enforced contract. Your server must accept this exact format.

inference_params: InferenceParams | None

instance: Dict[str, Any]

metadata: RolloutMetadata

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

model_endpoint: str | None

model_name: str | None