[Bug] Pipeline parameters (ParameterInteger, ParameterString) fail in ModelTrainer hyperparameters due to safe_serialize

[lopezfelipe]

PySDK Version

• PySDK V2 (2.x)
• PySDK V3 (3.x)

Describe the bug
Pipeline parameters (e.g., ParameterInteger, ParameterString) cannot be used in ModelTrainer hyperparameters because the safe_serialize function in sagemaker-train/src/sagemaker/train/utils.py doesn't handle PipelineVariable objects, causing a TypeError when building the pipeline.

To reproduce
Trying to pass max_depth as a Pipeline parameter to an XGBoost container in ModelTrainer

from sagemaker.core.workflow.parameters import ParameterInteger
from sagemaker.train import ModelTrainer
from sagemaker.core.training.configs import Compute
from sagemaker.core.workflow.pipeline_context import PipelineSession
from sagemaker.mlops.workflow.steps import TrainingStep
from sagemaker.mlops.workflow.pipeline import Pipeline

# Create pipeline session and parameter
pipeline_session = PipelineSession()
max_depth = ParameterInteger(name="MaxDepth", default_value=5)

# Create ModelTrainer with pipeline parameter in hyperparameters
model_trainer = ModelTrainer(
    training_image="683313688378.dkr.ecr.us-east-1.amazonaws.com/sagemaker-xgboost:1.0-1-cpu-py3",
    compute=Compute(instance_type="ml.m5.xlarge", instance_count=1),
    sagemaker_session=pipeline_session,
    role=role,
    hyperparameters={
        "max_depth": max_depth,  # Pipeline parameter
    },
)

train_args = model_trainer.train()
step_train = TrainingStep(name="TrainStep", step_args=train_args)

# Create and upsert pipeline
pipeline = Pipeline(
    name="test-pipeline",
    parameters=[max_depth],
    steps=[step_train],
    sagemaker_session=pipeline_session,
)

It will fail on this step:

pipeline.upsert(role_arn=role)

Expected behavior
Pipeline parameters should be serialized correctly and the pipeline should be created successfully, allowing hyperparameters to be parameterized at pipeline execution time.

Screenshots or logs
Error logs

│ /opt/conda/lib/python3.12/site-packages/sagemaker/train/utils.py:191 in safe_serialize           │
│                                                                                                  │
│   188 │   try:                                                                                   │
│   189 │   │   return json.dumps(data)                                                            │
│   190 │   except TypeError:                                                                      │
│ ❱ 191 │   │   return str(data)                                                                   │
│   192                                                                                            │
│   193                                                                                            │
│   194 def _run_clone_command_silent(repo_url, dest_dir):                                         │
│                                                                                                  │
│ /opt/conda/lib/python3.12/site-packages/sagemaker/core/helper/pipeline_variable.py:38 in __str__ │
│                                                                                                  │
│   35 │                                                                                           │
│   36 │   def __str__(self):                                                                      │
│   37 │   │   """Override built-in String function for PipelineVariable"""                        │
│ ❱ 38 │   │   raise TypeError(                                                                    │
│   39 │   │   │   "Pipeline variables do not support __str__ operation. "                         │
│   40 │   │   │   "Please use `.to_string()` to convert it to string type in execution time "     │
│   41 │   │   │   "or use `.expr` to translate it to Json for display purpose in Python SDK."     │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
TypeError: Pipeline variables do not support __str__ operation. Please use `.to_string()` to convert it to string 
type in execution time or use `.expr` to translate it to Json for display purpose in Python SDK.

System information
A description of your system. Please provide:

• SageMaker Python SDK version: 3.3.1
• Framework name (eg. PyTorch) or algorithm (eg. KMeans): XGBoost
• Framework version: 1.0
• Python version: 3.12.9
• CPU or GPU: CPU
• Custom Docker image (Y/N): N

Additional context
Add any other context about the problem here.

[ravees]

+1