sdgym 0.15.1.dev0__tar.gz → 0.15.2.dev0__tar.gz

{sdgym-0.15.1.dev0/sdgym.egg-info → sdgym-0.15.2.dev0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sdgym
-Version: 0.15.1.dev0
+Version: 0.15.2.dev0
 Summary: Benchmark tabular synthetic data generators using a variety of datasets
 Author-email: "DataCebo, Inc." <info@sdv.dev>
 License-Expression: BUSL-1.1
@@ -14,14 +14,13 @@ Classifier: Development Status :: 2 - Pre-Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Natural Language :: English
 Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
-Requires-Python: <3.15,>=3.9
+Requires-Python: <3.15,>=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: appdirs>=1.3
@@ -67,22 +66,31 @@ Requires-Dist: XlsxWriter>=1.2.8
 Requires-Dist: rdt>=1.18.2; python_version < "3.14"
 Requires-Dist: rdt>=1.20.0; python_version >= "3.14"
 Requires-Dist: sdmetrics>=0.29.0
-Requires-Dist: sdv>=1.37.0
+Requires-Dist: sdv<2,>=1.37.0
 Provides-Extra: dask
 Requires-Dist: dask; extra == "dask"
 Requires-Dist: distributed; extra == "dask"
 Provides-Extra: realtabformer
 Requires-Dist: realtabformer!=0.2.4,>=0.2.3; extra == "realtabformer"
 Requires-Dist: torch>=2.6.0; extra == "realtabformer"
-Requires-Dist: transformers<4.51; extra == "realtabformer"
 Provides-Extra: clavaddpm
 Requires-Dist: faiss-cpu>=1.10.0; python_version < "3.14" and extra == "clavaddpm"
 Requires-Dist: faiss-cpu>=1.12.0; python_version >= "3.14" and extra == "clavaddpm"
 Requires-Dist: scipy>=1.9.2; extra == "clavaddpm"
 Requires-Dist: numpy>=1.25.0; extra == "clavaddpm"
+Provides-Extra: data-designer
+Requires-Dist: data-designer>=0.8; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: google-api-core<2.36; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: pandas>=2.3.3; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: scipy>=1.11.0; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: transformers>=5.0.0; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: boto3>=1.34.63; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: botocore>=1.34.63; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: pyOpenSSL>=26.4.0; python_version >= "3.10" and extra == "data-designer"
 Provides-Extra: test
 Requires-Dist: sdgym[realtabformer]; extra == "test"
 Requires-Dist: sdgym[clavaddpm]; extra == "test"
+Requires-Dist: sdgym[data_designer]; extra == "test"
 Requires-Dist: pytest>=6.2.5; extra == "test"
 Requires-Dist: pytest-cov>=2.6.0; extra == "test"
 Requires-Dist: jupyter<2,>=1.0.0; extra == "test"

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/pyproject.toml

@@ -7,7 +7,6 @@ classifiers = [
     'Intended Audience :: Developers',
     'Natural Language :: English',
     'Programming Language :: Python :: 3',
-    'Programming Language :: Python :: 3.9',
     'Programming Language :: Python :: 3.10',
     'Programming Language :: Python :: 3.11',
     'Programming Language :: Python :: 3.12',
@@ -19,7 +18,7 @@ keywords = ['machine learning', 'synthetic data generation', 'benchmark', 'gener
 dynamic = ['version']
 license = 'BUSL-1.1'
 license-files = ['LICENSE']
-requires-python = '>=3.9,<3.15'
+requires-python = '>=3.10,<3.15'
 readme = 'README.md'
 dependencies = [
     'appdirs>=1.3',
@@ -65,7 +64,7 @@ dependencies = [
     "rdt>=1.18.2;python_version<'3.14'",
     "rdt>=1.20.0;python_version>='3.14'",
     "sdmetrics>=0.29.0",
-    "sdv>=1.37.0",
+    "sdv>=1.37.0,<2",
 ]
 
 [project.urls]
@@ -83,7 +82,6 @@ dask = ['dask', 'distributed']
 realtabformer = [
     'realtabformer>=0.2.3,!=0.2.4',
     "torch>=2.6.0",
-    'transformers<4.51',
 ]
 clavaddpm = [
     "faiss-cpu>=1.10.0;python_version<'3.14'",
@@ -91,9 +89,20 @@ clavaddpm = [
     'scipy>=1.9.2',
     'numpy>=1.25.0',
 ]
+data_designer = [
+    "data-designer>=0.8;python_version>='3.10'",
+    "google-api-core<2.36;python_version>='3.10'",  # requires opentelemetry-api<1.44
+    "pandas>=2.3.3;python_version>='3.10'",
+    "scipy>=1.11.0;python_version>='3.10'",
+    "transformers>=5.0.0;python_version>='3.10'",  # requires huggingface-hub>=1
+    "boto3>=1.34.63;python_version>='3.10'",  # requires urllib3>=2.7
+    "botocore>=1.34.63;python_version>='3.10'",
+    "pyOpenSSL>=26.4.0;python_version>='3.10'",  # requires cryptography>=48
+]
 test = [
     'sdgym[realtabformer]',
     'sdgym[clavaddpm]',
+    'sdgym[data_designer]',
     'pytest>=6.2.5',
     'pytest-cov>=2.6.0',
     'jupyter>=1.0.0,<2',
@@ -168,7 +177,7 @@ namespaces = false
 version = {attr = 'sdgym.__version__'}
 
 [tool.bumpversion]
-current_version = "0.15.1.dev0"
+current_version = "0.15.2.dev0"
 parse = '(?P<major>\d+)\.(?P<minor>\d+)\.(?P<patch>\d+)(\.(?P<release>[a-z]+)(?P<candidate>\d+))?'
 serialize = [
     '{major}.{minor}.{patch}.{release}{candidate}',

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym/__init__.py

@@ -8,7 +8,7 @@ __author__ = 'DataCebo, Inc.'
 __copyright__ = 'Copyright (c) 2022 DataCebo, Inc.'
 __email__ = 'info@sdv.dev'
 __license__ = 'BSL-1.1'
-__version__ = '0.15.1.dev0'
+__version__ = '0.15.2.dev0'
 
 import logging
 

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym/_benchmark_launcher/utils.py

@@ -43,6 +43,7 @@ MODALITY_TO_JOB_SETUP = {
             'BootstrapSynthesizer',
             'RealTabFormerSynthesizer',
             'TabDDPMSynthesizer',
+            'DataDesignerSynthesizer',
         ],
     },
     'multi_table': {

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym/synthesizer_descriptions.yaml

@@ -134,7 +134,7 @@ XGCSynthesizer:
     to train a model and generate synthetic data similar to the GaussianCopulaSynthesizer.
     However, it contains some additional features for higher quality modeling.
 
-TabDDOMSynthesizer:
+TabDDPMSynthesizer:
   library: sdgym
   organization: Yandex
   modality: single_table
@@ -152,4 +152,13 @@ ClavaDDPMSynthesizer:
   description: >
     The ClavaDDPM model is an extension of the TabDDPM model to handle multi-table datasets.
     It is based on clustering labels as intermediaries to model relationships between tables,
-    specifically focusing on foreign key constraints.
+    specifically focusing on foreign key constraints.
+
+DataDesignerSynthesizer:
+  library: data-designer
+  organization: Nvidia
+  modality: single_table
+  type: Statistical and Large Language Models
+  description: >
+    NVIDIA Data Designer is a configurable framework that uses LLMs, statistical models,
+    and customizable samplers to generate synthetic data.

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym/synthesizers/__init__.py

@@ -7,6 +7,7 @@ from sdgym.synthesizers.generate import (
 )
 from sdgym.synthesizers.identity import DataIdentity
 from sdgym.synthesizers.column import ColumnSynthesizer
+from sdgym.synthesizers.data_designer import DataDesignerSynthesizer
 from sdgym.synthesizers.realtabformer import RealTabFormerSynthesizer
 from sdgym.synthesizers.tabddpm import TabDDPMSynthesizer
 from sdgym.synthesizers.clavaddpm import ClavaDDPMSynthesizer
@@ -21,6 +22,7 @@ from sdgym.synthesizers.sdv import create_sdv_synthesizer_class, _get_all_sdv_sy
 __all__ = [
     'DataIdentity',
     'ColumnSynthesizer',
+    'DataDesignerSynthesizer',
     'UniformSynthesizer',
     'RealTabFormerSynthesizer',
     'TabDDPMSynthesizer',

sdgym-0.15.2.dev0/sdgym/synthesizers/data_designer.py

@@ -0,0 +1,346 @@
+"""NVIDIA Data Designer integration."""
+
+import logging
+import shutil
+import sys
+import tempfile
+import uuid
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+from sdv.metadata import Metadata
+
+from sdgym.synthesizers.base import BaselineSynthesizer
+
+LOGGER = logging.getLogger(__name__)
+
+MIN_PYTHON_VERSION = (3, 10)
+DEFAULT_MODEL_ALIAS = 'text'
+DEFAULT_MODEL_ID = 'nvidia/nvidia-nemotron-nano-9b-v2'
+DEFAULT_MODEL_PROVIDER = 'nvidia'
+DEFAULT_TEMPERATURE = 0.85
+DEFAULT_TOP_P = 0.95
+PII_PLACEHOLDER_PREFIX = 'sdgym-pii-'
+MAX_TEXT_EXAMPLES = 3
+MAX_TEXT_EXAMPLE_LENGTH = 100
+
+MISSING_PROPORTION_ATTRIBUTES = (
+    'missing_values_proportion',
+    'missing_proportion',
+    'null_proportion',
+)
+# sdtypes that can be useful as context
+TEXT_CONTEXT_SDTYPES = {'numerical', 'categorical', 'boolean', 'datetime'}
+
+
+def _import_data_designer():
+    if sys.version_info < MIN_PYTHON_VERSION:
+        raise ImportError('DataDesignerSynthesizer only supports python >= 3.10')
+
+    try:
+        import data_designer.config as data_designer_config
+    except ImportError as exception:
+        raise ImportError(
+            "To use 'DataDesignerSynthesizer' you have to install the extra "
+            "dependencies by running pip install sdgym['data_designer']"
+        ) from exception
+
+    return data_designer_config
+
+
+def _get_table_metadata(metadata, table_name=None):
+    """Return the metadata of ``table_name`` from a ``Metadata`` object."""
+    if not isinstance(metadata, Metadata):
+        raise TypeError(f'Expected sdv.Metadata, got {type(metadata).__name__}. ')
+
+    if table_name is None:
+        if len(metadata.tables) != 1:
+            raise ValueError(
+                f'Metadata has {len(metadata.tables)} tables, please provide a `table_name`. '
+                'DataDesignerSynthesizer only supports single table metadata.'
+            )
+
+        table_name = metadata._get_single_table_name()
+
+    if table_name not in metadata.tables:
+        raise ValueError(f"Table '{table_name}' is not present in the metadata.")
+
+    return metadata.tables[table_name]
+
+
+def _create_default_model_config(
+    dd,
+    model_alias,
+    model=DEFAULT_MODEL_ID,
+    provider=DEFAULT_MODEL_PROVIDER,
+    temperature=DEFAULT_TEMPERATURE,
+    top_p=DEFAULT_TOP_P,
+):
+
+    return dd.ModelConfig(
+        alias=model_alias,
+        model=model,
+        provider=provider,
+        inference_parameters=dd.ChatCompletionInferenceParams(temperature=temperature, top_p=top_p),
+    )
+
+
+def _create_categorical_config(dd, column_name, column_data, column_metadata):
+    """Map a categorical column to a weighted ``category`` sampler."""
+    value_counts = column_data.dropna().value_counts()
+    values = column_metadata.get('range_values')
+    if values is None:
+        values = list(value_counts.index)
+
+    if not values:
+        LOGGER.warning(
+            f"Column '{column_name}' has no observed values, using a placeholder sampler."
+        )
+        return _create_placeholder_config(dd, column_name)
+
+    values = [value.item() if isinstance(value, np.generic) else value for value in values]
+    weights = [float(value_counts.get(value, 0)) for value in values]
+    params = dd.CategorySamplerParams(values=values, weights=weights if sum(weights) else None)
+    return dd.SamplerColumnConfig(
+        name=column_name, sampler_type=dd.SamplerType.CATEGORY, params=params
+    )
+
+
+def _create_boolean_config(dd, column_name, column_data):
+    """Map a boolean column to a ``bernoulli`` sampler."""
+    values = column_data.dropna()
+    probability = float(values.astype(bool).mean()) if not values.empty else 0.5
+    params = dd.BernoulliSamplerParams(p=probability)
+    return dd.SamplerColumnConfig(
+        name=column_name, sampler_type=dd.SamplerType.BERNOULLI, params=params
+    )
+
+
+def _get_decimal_places(values):
+    """Return the largest number of decimal places observed in ``values``."""
+    decimals = 0
+    for value in values.head(1000):
+        _, _, fraction = repr(float(value)).partition('.')
+        if fraction and fraction != '0':
+            decimals = max(decimals, len(fraction))
+
+    return decimals
+
+
+def _create_numerical_config(dd, column_name, column_data, column_metadata):
+    """Map a numerical column to a ``uniform`` sampler."""
+    values = pd.to_numeric(column_data, errors='coerce').dropna()
+    values_empty = values.empty
+    has_range = 'range_min' in column_metadata and 'range_max' in column_metadata
+    if values_empty and not has_range:
+        LOGGER.warning(
+            f"Column '{column_name}' has no numeric values, sampling from normal distribution."
+        )
+        return dd.SamplerColumnConfig(
+            name=column_name,
+            sampler_type=dd.SamplerType.GAUSSIAN,
+            params=dd.GaussianSamplerParams(mean=0, stddev=1),
+        )
+
+    low = column_metadata.get('range_min', values.min())
+    high = column_metadata.get('range_max', values.max())
+    decimal_places = column_metadata.get('decimal_places', _get_decimal_places(values))
+    is_integer = decimal_places == 0
+
+    params = dd.UniformSamplerParams(
+        low=float(low), high=float(high), decimal_places=decimal_places
+    )
+    return dd.SamplerColumnConfig(
+        name=column_name,
+        sampler_type=dd.SamplerType.UNIFORM,
+        params=params,
+        convert_to='int' if is_integer else None,
+    )
+
+
+def _create_datetime_config(dd, column_name, column_data, column_metadata):
+    """Map a datetime column to a ``datetime`` sampler."""
+    datetime_format = column_metadata.get('datetime_format')
+    values = pd.to_datetime(column_data, format=datetime_format, errors='coerce').dropna()
+    start = pd.to_datetime(column_metadata.get('range_min', values.min()), format=datetime_format)
+    end = pd.to_datetime(column_metadata.get('range_max', values.max()), format=datetime_format)
+
+    if pd.isna(start) and pd.isna(end):
+        LOGGER.warning(
+            f"Column '{column_name}' has no parseable datetimes, using a placeholder sampler."
+        )
+        return _create_placeholder_config(dd, column_name)
+
+    elif pd.isna(start):
+        start = end
+    elif pd.isna(end):
+        end = start
+
+    output_format = datetime_format or '%Y-%m-%d %H:%M:%S'
+    params = dd.DatetimeSamplerParams(
+        start=start.strftime(output_format), end=end.strftime(output_format)
+    )
+    return dd.SamplerColumnConfig(
+        name=column_name, sampler_type=dd.SamplerType.DATETIME, params=params
+    )
+
+
+def _create_id_config(dd, column_name):
+    """Map an id column to a ``uuid`` sampler, no other id sampler is available."""
+    return dd.SamplerColumnConfig(
+        name=column_name, sampler_type=dd.SamplerType.UUID, params=dd.UUIDSamplerParams()
+    )
+
+
+def _create_placeholder_config(dd, column_name):
+    """Map unknown or unsupported pii columns to short anonymized placeholders."""
+    params = dd.UUIDSamplerParams(prefix=PII_PLACEHOLDER_PREFIX, short_form=True)
+    return dd.SamplerColumnConfig(name=column_name, sampler_type=dd.SamplerType.UUID, params=params)
+
+
+def _create_text_config(dd, column_name, column_data, model_alias, context_columns):
+    """Map a text column to an LLM generated column conditioned on the rest of the row."""
+    examples = []
+    for value in column_data.dropna().astype(str).unique()[:MAX_TEXT_EXAMPLES]:
+        examples.append(repr(value[:MAX_TEXT_EXAMPLE_LENGTH]))
+
+    prompt = (
+        'You are a helpful assistant that generates synthetic data.'
+        f"Generate a realistic value for the column '{column_name}' of a tabular dataset."
+    )
+    if examples:
+        prompt += f' Here are a few examples from this column: {", ".join(examples)}.'
+
+    if context_columns:
+        row = ', '.join(f'{{ {name} }}' for name in context_columns)
+        prompt += f' Refer to the values in these columns for context: {row}.'
+
+    prompt += ' Respond with the value only, without any explanation.'
+    return dd.LLMTextColumnConfig(name=column_name, prompt=prompt, model_alias=model_alias)
+
+
+def create_data_designer_config(
+    data,
+    metadata,
+    table_name=None,
+    model_alias=DEFAULT_MODEL_ALIAS,
+    model=DEFAULT_MODEL_ID,
+    provider=DEFAULT_MODEL_PROVIDER,
+    temperature=DEFAULT_TEMPERATURE,
+    top_p=DEFAULT_TOP_P,
+):
+    """Map a single table dataset and its metadata to a Data Designer config builder.
+
+    Args:
+        data (pd.DataFrame):
+            The real data. Used to fit the sampler parameters that the metadata does not
+            provide (frequencies, ranges, boolean rates) and to pick text examples.
+        metadata (sdv.Metadata):
+            Metadata describing the data. If range values are not present, they are
+            extracted directly from data.
+        table_name (str or None):
+            The table to map when the metadata has more than one table. Defaults to the only
+            table.
+        model_alias (str):
+            The model alias used by LLM generated text columns. Defaults to 'text'.
+        model (str):
+            The model identifier used for inference. This must correspond to the
+            served model name configured by the specified provider.
+        provider (str):
+            The provider identifier used to route inference requests to the specified model.
+        temperature (float):
+            Sampling temperature of the model used by LLM generated text columns. Higher
+            values produce more varied text. Defaults to 0.85.
+        top_p (float):
+            Nucleus sampling probability of the model used by LLM generated text columns.
+            Defaults to 0.95.
+
+    Returns:
+        DataDesignerConfigBuilder:
+            A builder holding one column config per metadata column.
+    """
+    dd = _import_data_designer()
+    table_metadata = _get_table_metadata(metadata, table_name)
+    model_configs = [
+        _create_default_model_config(dd, model_alias, model, provider, temperature, top_p)
+    ]
+
+    context_columns = [
+        name
+        for name, column_meta in table_metadata.columns.items()
+        if column_meta['sdtype'] in TEXT_CONTEXT_SDTYPES
+    ]
+
+    column_configs = []
+    for column_name, column_metadata in table_metadata.columns.items():
+        sdtype = column_metadata['sdtype']
+        column_data = data[column_name]
+        if sdtype == 'boolean':
+            config = _create_boolean_config(dd, column_name, column_data)
+        elif sdtype == 'numerical':
+            config = _create_numerical_config(dd, column_name, column_data, column_metadata)
+        elif sdtype == 'datetime':
+            config = _create_datetime_config(dd, column_name, column_data, column_metadata)
+        elif sdtype == 'id':
+            config = _create_id_config(dd, column_name)
+        elif sdtype == 'text':
+            config = _create_text_config(dd, column_name, column_data, model_alias, context_columns)
+        else:
+            config = _create_categorical_config(dd, column_name, column_data, column_metadata)
+
+        column_configs.append(config)
+
+    builder = dd.DataDesignerConfigBuilder(model_configs=model_configs)
+    for config in column_configs:
+        builder.add_column(config)
+
+    return builder
+
+
+class DataDesignerSynthesizer(BaselineSynthesizer):
+    """Custom wrapper for the DataDesigner synthesizer to make it work with SDGym."""
+
+    LOGGER = logging.getLogger(__name__)
+    _MODEL_KWARGS = None
+    _MODALITY_FLAG = 'single_table'
+
+    def _fit(self, data, metadata):
+        table_name = metadata._get_single_table_name()
+        metadata.validate()
+        metadata.validate_data({table_name: data})
+
+        model_kwargs = self._MODEL_KWARGS.copy() if self._MODEL_KWARGS else {}
+        self._artifact_path = model_kwargs.pop('artifact_path', None)
+        self._cleanup_artifacts = model_kwargs.pop('cleanup_artifacts', True)
+        self._config_builder = create_data_designer_config(data, metadata, **model_kwargs)
+
+    def _sample_from_synthesizer(self, synthesizer, n_sample):
+        """Sample synthetic data with specified sample count."""
+        try:
+            from data_designer.interface import DataDesigner
+        except Exception as exception:
+            raise ImportError(
+                "To use 'DataDesignerSynthesizer' you have to install the extra "
+                "dependencies by running pip install sdgym['data_designer']"
+            ) from exception
+
+        is_temporary = synthesizer._artifact_path is None
+        if is_temporary:
+            artifact_path = Path(tempfile.mkdtemp(prefix='sdgym-data-designer-'))
+        else:
+            artifact_path = Path(synthesizer._artifact_path)
+
+        # A unique name keeps concurrent or repeated sample calls from sharing a folder.
+        dataset_name = f'dataset-{uuid.uuid4().hex}'
+        try:
+            output = DataDesigner(artifact_path=artifact_path).create(
+                synthesizer._config_builder, num_records=n_sample, dataset_name=dataset_name
+            )
+            return output.load_dataset()
+        finally:
+            if synthesizer._cleanup_artifacts:
+                to_remove = artifact_path if is_temporary else artifact_path / dataset_name
+                shutil.rmtree(to_remove, ignore_errors=True)
+            else:
+                self.LOGGER.info(f'Data Designer artifacts kept in {artifact_path / dataset_name}')

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym/synthesizers/realtabformer.py

@@ -1,6 +1,7 @@
 """REaLTabFormer integration."""
 
 import contextlib
+import dataclasses
 import logging
 from functools import partialmethod
 
@@ -8,6 +9,8 @@ import tqdm
 
 from sdgym.synthesizers.base import BaselineSynthesizer
 
+REMOVED_TRAINING_ARGS = ('overwrite_output_dir',)
+
 
 @contextlib.contextmanager
 def prevent_tqdm_output():
@@ -19,6 +22,16 @@ def prevent_tqdm_output():
         tqdm.__init__ = partialmethod(tqdm.__init__, disable=False)
 
 
+def _remove_unsupported_training_args(model):
+    """Drop deprecated args that are no longer used by transformers."""
+    from transformers import TrainingArguments
+
+    supported = {field.name for field in dataclasses.fields(TrainingArguments)}
+    for name in REMOVED_TRAINING_ARGS:
+        if name not in supported:
+            model.training_args_kwargs.pop(name, None)
+
+
 class RealTabFormerSynthesizer(BaselineSynthesizer):
     """Custom wrapper for the REaLTabFormer synthesizer to make it work with SDGym."""
 
@@ -39,6 +52,7 @@ class RealTabFormerSynthesizer(BaselineSynthesizer):
         with prevent_tqdm_output():
             model_kwargs = self._MODEL_KWARGS.copy() if self._MODEL_KWARGS else {}
             model = REaLTabFormer(model_type='tabular', **model_kwargs)
+            _remove_unsupported_training_args(model)
             model.fit(data)
 
         self._internal_synthesizer = model

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0/sdgym.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sdgym
-Version: 0.15.1.dev0
+Version: 0.15.2.dev0
 Summary: Benchmark tabular synthetic data generators using a variety of datasets
 Author-email: "DataCebo, Inc." <info@sdv.dev>
 License-Expression: BUSL-1.1
@@ -14,14 +14,13 @@ Classifier: Development Status :: 2 - Pre-Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Natural Language :: English
 Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
-Requires-Python: <3.15,>=3.9
+Requires-Python: <3.15,>=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: appdirs>=1.3
@@ -67,22 +66,31 @@ Requires-Dist: XlsxWriter>=1.2.8
 Requires-Dist: rdt>=1.18.2; python_version < "3.14"
 Requires-Dist: rdt>=1.20.0; python_version >= "3.14"
 Requires-Dist: sdmetrics>=0.29.0
-Requires-Dist: sdv>=1.37.0
+Requires-Dist: sdv<2,>=1.37.0
 Provides-Extra: dask
 Requires-Dist: dask; extra == "dask"
 Requires-Dist: distributed; extra == "dask"
 Provides-Extra: realtabformer
 Requires-Dist: realtabformer!=0.2.4,>=0.2.3; extra == "realtabformer"
 Requires-Dist: torch>=2.6.0; extra == "realtabformer"
-Requires-Dist: transformers<4.51; extra == "realtabformer"
 Provides-Extra: clavaddpm
 Requires-Dist: faiss-cpu>=1.10.0; python_version < "3.14" and extra == "clavaddpm"
 Requires-Dist: faiss-cpu>=1.12.0; python_version >= "3.14" and extra == "clavaddpm"
 Requires-Dist: scipy>=1.9.2; extra == "clavaddpm"
 Requires-Dist: numpy>=1.25.0; extra == "clavaddpm"
+Provides-Extra: data-designer
+Requires-Dist: data-designer>=0.8; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: google-api-core<2.36; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: pandas>=2.3.3; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: scipy>=1.11.0; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: transformers>=5.0.0; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: boto3>=1.34.63; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: botocore>=1.34.63; python_version >= "3.10" and extra == "data-designer"
+Requires-Dist: pyOpenSSL>=26.4.0; python_version >= "3.10" and extra == "data-designer"
 Provides-Extra: test
 Requires-Dist: sdgym[realtabformer]; extra == "test"
 Requires-Dist: sdgym[clavaddpm]; extra == "test"
+Requires-Dist: sdgym[data_designer]; extra == "test"
 Requires-Dist: pytest>=6.2.5; extra == "test"
 Requires-Dist: pytest-cov>=2.6.0; extra == "test"
 Requires-Dist: jupyter<2,>=1.0.0; extra == "test"

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym.egg-info/SOURCES.txt

@@ -47,6 +47,7 @@ sdgym/synthesizers/__init__.py
 sdgym/synthesizers/base.py
 sdgym/synthesizers/clavaddpm.py
 sdgym/synthesizers/column.py
+sdgym/synthesizers/data_designer.py
 sdgym/synthesizers/generate.py
 sdgym/synthesizers/identity.py
 sdgym/synthesizers/realtabformer.py

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/sdgym.egg-info/requires.txt

@@ -11,7 +11,7 @@ tabulate<0.9,>=0.8.3
 tqdm>=4.66.3
 XlsxWriter>=1.2.8
 sdmetrics>=0.29.0
-sdv>=1.37.0
+sdv<2,>=1.37.0
 
 [:python_version < "3.10"]
 numpy>=1.22.2
@@ -78,6 +78,18 @@ faiss-cpu>=1.12.0
 dask
 distributed
 
+[data_designer]
+
+[data_designer:python_version >= "3.10"]
+data-designer>=0.8
+google-api-core<2.36
+pandas>=2.3.3
+scipy>=1.11.0
+transformers>=5.0.0
+boto3>=1.34.63
+botocore>=1.34.63
+pyOpenSSL>=26.4.0
+
 [dev]
 sdgym[dask,test]
 build<2,>=1.0.0
@@ -97,11 +109,11 @@ rundoc<0.5,>=0.4.3
 [realtabformer]
 realtabformer!=0.2.4,>=0.2.3
 torch>=2.6.0
-transformers<4.51
 
 [test]
 sdgym[realtabformer]
 sdgym[clavaddpm]
+sdgym[data_designer]
 pytest>=6.2.5
 pytest-cov>=2.6.0
 jupyter<2,>=1.0.0

{sdgym-0.15.1.dev0 → sdgym-0.15.2.dev0}/tests/test_tasks.py

@@ -121,7 +121,7 @@ def _get_example_pyproject_dict():
                 ],
             },
             'readme': 'README.md',
-            'requires-python': '>=3.9,<3.15',
+            'requires-python': '>=3.10,<3.15',
         },
         'tool': {
             'bumpversion': {