cursus.steps.configs.config_sopa_instruction_tuning_step¶

SOPA Instruction Tuning Step Configuration with Self-Contained Derivation Logic

This module implements the configuration class for SageMaker SOPA (Stage Of Pre-training Alignment) Stage 2 instruction fine-tuning steps using a self-contained design where derived fields are private with read-only properties.

Fields are organized into three tiers: 1. Tier 1: Essential User Inputs - fields that users must explicitly provide 2. Tier 2: System Inputs with Defaults - fields with reasonable defaults that can be overridden 3. Tier 3: Derived Fields - fields calculated from other fields (private with properties)

Note: Most training/model hyperparameters (feature_columns, input_size, latent_size, batch_size, lr, epochs, etc.) are defined in SOPAInstructionTuningHyperparameters and serialized to a JSON file. They are NOT duplicated here. The builder loads them from the JSON at pipeline construction time.

Only infrastructure settings, filenames, and the task selector live in this config.

class SOPAInstructionTuningConfig(*, author, bucket, role, region, service_name, pipeline_version, model_class='xgboost', current_date=<factory>, framework_version='2.1.0', py_version='py310', source_dir=None, enable_caching=False, use_secure_pypi=False, max_runtime_seconds=172800, project_root_folder, training_entry_point, task, training_instance_type='ml.p4d.24xlarge', training_instance_count=1, training_volume_size=50, training_data_filename='training_data.csv', pretrained_qformer='qformer', tabular_encoder_filename='autoencoder_model', hyperparameters_dict=<factory>, **extra_data)[source]¶

Bases: BasePipelineConfig

Configuration specific to the SageMaker SOPA Instruction Tuning Step.

This configuration collects user inputs for SOPA Stage 2 instruction fine-tuning, which fine-tunes a BLIP2-based model (Q-Former + Phi-3 LLM) for tabular-to-text instruction following tasks. Input/output paths are provided via step specifications and dependencies.

Most training/model hyperparameters are defined separately in SOPAInstructionTuningHyperparameters and loaded from the hyperparameters JSON by the builder. This config only contains: - Infrastructure settings (instance type, count, volume) - Task selector - Filename overrides (training data, Q-Former checkpoint, tabular encoder checkpoint)

training_entry_point: str¶

task: str¶

training_instance_type: str¶

training_instance_count: int¶

training_volume_size: int¶

training_data_filename: str¶

pretrained_qformer: str¶

tabular_encoder_filename: str¶

hyperparameters_dict: Dict[str, Any]¶

model_config: ClassVar[ConfigDict] = {'arbitrary_types_allowed': True, 'extra': 'allow', 'protected_namespaces': (), 'validate_assignment': True}¶: Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

property hyperparameter_file: str¶: Get hyperparameter file path.

property checkpoint_s3_uri: str¶: Get S3 URI for model checkpoints.

model_dump(**kwargs)[source]¶

Override model_dump to include derived properties.

initialize_derived_fields()[source]¶

Initialize all derived fields once after validation.

classmethod validate_task(v)[source]¶

Validate task is one of allowed instruction tuning tasks.

get_environment_variables()[source]¶

Get environment variables for the SOPA instruction tuning script.

The SOPA script uses argparse for all configuration and does not require environment variables.

Returns:: Empty dictionary (no env vars per script contract)
Return type:: Dict[str, str]

get_public_init_fields()[source]¶

Override get_public_init_fields to include SOPA instruction tuning-specific fields. Gets a dictionary of public fields suitable for initializing a child config. Includes both base fields (from parent) and SOPA-specific fields.

Note: Training/model hyperparameters (feature_columns, input_size, batch_size, lr, etc.) are NOT included here — they come from the hyperparameters JSON file.

Returns:: Dictionary of field names to values for child initialization
Return type:: Dict[str, Any]

model_post_init(context, /)¶

This function is meant to behave like a BaseModel method to initialize private attributes.

It takes context as an argument since that’s what pydantic-core passes when calling it.

Parameters:

self (BaseModel) – The BaseModel instance.
context (Any) – The context.