does SageMaker Pipeline in SageMaker Python SDK v3 support fine-tuning (such as SFTTrainer, DPOTrainer, RLAIFTrainer)?

[philipskokoh]

PySDK Version

• PySDK V2 (2.x)
• PySDK V3 (3.x)

Describe the bug
Hi team, I have a quick question, does SageMaker Pipeline in SageMaker Python SDK v3 support fine-tuning (such as SFTTrainer, DPOTrainer, RLAIFTrainer)?

I have this kind of codes. I created SFTTrainer() with PipelineSession in the sagemaker_session. However, the train() is not defered, but automatially launches training job for fine-tuning.

To reproduce

session = PipelineSession()

sft_trainer = SFTTrainer(
    model="huggingface-reasoning-qwen3-1-7b",
    training_type=TrainingType.LORA,
    training_dataset=f"{TRIAGE_DATA_PREFIX}/anycompany_triage_sft.jsonl",
    model_package_group="anycompany-triage-agent", 
    s3_output_path=f"{PIPELINE_OUTPUT_PREFIX}/sft-output",
    sagemaker_session=session,
)

step_sft_args = sft_trainer.train()   # Expect to be deferred inside pipeline-definition context, but automatically run

step_sft = TrainingStep(
    name="SFTTuneTriage",
    step_args=step_sft_args,
)
...

Expected behavior
The function is not automatically run, but defined as a pipeline in SageMaker Pipeline.

System information
A description of your system. Please provide:

• SageMaker Python SDK version: 3.17.0