`safe_synthesizer.job_builder`#

Module Contents#

Classes#

SafeSynthesizerJobBuilder

Builder for Safe Synthesizer Jobs ran with the NeMo Microservices Platform.

Data#

logger

API#

class safe_synthesizer.job_builder.SafeSynthesizerJobBuilder( client: nemo_platform.NeMoPlatform, workspace: str = 'default', )#

Builder for Safe Synthesizer Jobs ran with the NeMo Microservices Platform.

This class provides a fluent interface for building Safe Synthesizer configurations. It allows you to configure all the parameters needed to create and run a Safe Synthesizer job. Each method returns the builder instance to allow method chaining.

Examples

>>> from nemo_platform import NeMoPlatform
>>> from nemo_platform.beta.safe_synthesizer.job_builder import SafeSynthesizerJobBuilder
>>> client = NeMoPlatform(base_url=..., inference_base_url=...)
>>> builder = (
...     SafeSynthesizerJobBuilder(client)
...     .with_data_source(your_dataframe)
...     .with_replace_pii()
...     .synthesize()
...     .with_train(learning_rate=0.0001)
...     .with_generate(num_records=10000)
...     .with_evaluate(enable=False)
... )
>>> job = builder.create_job()

Initialization

create_job(**kwargs) → safe_synthesizer.job.SafeSynthesizerJob#

Upload the dataset and submit the job.

Parameters:: **kwargs – Additional job creation parameters passed to the API.
Returns:: A SafeSynthesizerJob for monitoring and retrieving results.

resolve_job_config() → typing_extensions.Self#

Resolve and validate the final job configuration without submitting.

Returns:: The builder instance for method chaining.

synthesize() → typing_extensions.Self#: Enable data synthesis for the job run.

with_classify_model_provider( provider_name: str, ) → typing_extensions.Self#

Configure column classification using an Inference Gateway model provider.

The model provider should be configured to serve an LLM suitable for column classification tasks.

Parameters:: provider_name – Name of the model provider. Provide just the name to use a provider in the current workspace (e.g. "my-classify-llm"), or a fully-qualified workspace/provider_name reference to use a provider from a different workspace.
Returns:: The builder instance for method chaining.

with_data(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#: Configure data parameters.

with_data_source( df_source: pandas.DataFrame | str, ) → typing_extensions.Self#

Set the data source for synthetic data generation.

Parameters:: df_source – Training dataset as a pandas DataFrame or a fetchable URL.
Returns:: The builder instance for method chaining.

with_differential_privacy(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#: Configure differential privacy parameters.

with_evaluate(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#: Configure evaluation parameters.

with_generate(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#

Configure generation parameters.

Calling this method also enables synthesis.

with_hf_token_secret(secret_name: str) → typing_extensions.Self#

Configure HuggingFace authentication using a platform secret.

The secret must exist in the same workspace as the job and should contain a valid HuggingFace token.

Parameters:: secret_name – Name of the platform secret containing the HuggingFace token.
Returns:: The builder instance for method chaining.

with_replace_pii(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#

Configure PII replacement.

Calling this method enables PII replacement for the job. If no config is provided the service will use its own defaults.

Parameters:

config – PII replacement config as a dict or Pydantic model, or None to use service-side defaults.
**kwargs – Individual PII replacement parameters to override.

Returns:

The builder instance for method chaining.

with_time_series(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#: Configure time-series parameters.

with_train(

config: safe_synthesizer.job_builder._ConfigInput = None,

**kwargs,

) → typing_extensions.Self#

Configure training hyperparameters.