model customization v3 release PR

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## v3.1.0 (2025-12-03)
+
+### Features
+
+* Fine-tuning SDK: SFT, RLVR, and RLAIF techniques with standardized parameter design
+* AIRegistry Integration: Added CRUD operations for datasets and evaluators
+* Enhanced Training Experience: Implemented MLFlow metrics tracking and deployment workflows
+
 ## v3.0.1 (2025-11-19)
 
 * Update project dependencies to include submodules: sagemaker-core, sagemaker-train, sagemaker-serve, sagemaker-mlops

VERSION

@@ -1 +1 @@
-3.0.1
+3.1.0

pyproject.toml

@@ -33,10 +33,10 @@ classifiers = [
   "Programming Language :: Python :: 3.12",
 ]
 dependencies = [
-  "sagemaker-core>=2.0.0,<3.0.0",
-  "sagemaker-train<2.0.0",
-  "sagemaker-serve<2.0.0",
-  "sagemaker-mlops<2.0.0",
+  "sagemaker-core>=2.1.0,<3.0.0",
+  "sagemaker-train>=1.1.0,<2.0.0",
+  "sagemaker-serve>=1.1.0,<2.0.0",
+  "sagemaker-mlops>=1.1.0,<2.0.0",
 ]
 
 [project.optional-dependencies]
@@ -59,4 +59,4 @@ addopts = ["-vv"]
 testpaths = ["tests"]
 
 [tool.black]
-line-length = 100
+line-length = 100

sagemaker-core/VERSION

@@ -1 +1,2 @@
-2.0.1
+2.1.0
+

sagemaker-core/pyproject.toml

@@ -12,7 +12,7 @@ authors = [
 readme = "README.rst"
 dependencies = [
   # Add your dependencies here (Include lower and upper bounds as applicable)
-    "boto3>=1.35.75,<2.0.0",
+    "boto3>=1.42.2,<2.0.0",
     "pydantic>=2.0.0,<3.0.0",
     "PyYAML>=6.0, <7.0",
     "jsonschema<5.0.0",

sagemaker-core/src/sagemaker/core/__init__.py

@@ -11,5 +11,6 @@
     FrameworkProcessor,
 )
 from sagemaker.core.transformer import Transformer  # noqa: F401
+
 # Note: HyperparameterTuner and WarmStartTypes are in sagemaker.train.tuner
 # They are not re-exported from core to avoid circular dependencies

sagemaker-core/src/sagemaker/core/_studio.py

@@ -113,4 +113,4 @@ def _parse_tags(config):
             {"Key": "sagemaker:project-name", "Value": config["sagemakerProjectName"]},
         ]
     except Exception as e:  # pylint: disable=W0703
-        logger.debug("Could not parse project config. %s", e)
+        logger.debug("Could not parse project config. %s", e)

sagemaker-core/src/sagemaker/core/apiutils/_base_types.py

@@ -225,4 +225,4 @@ def _invoke_api(self, boto_method, boto_method_members):
         api_kwargs = self.to_boto(api_values)
         api_method = getattr(self.sagemaker_session.sagemaker_client, boto_method)
         api_boto_response = api_method(**api_kwargs)
-        return self.with_boto(api_boto_response)
+        return self.with_boto(api_boto_response)

sagemaker-core/src/sagemaker/core/apiutils/_boto_functions.py

@@ -37,7 +37,7 @@ def to_pascal_case(snake_case):
     Returns:
         str: String converted to PascalCase.
     """
-    return ''.join(word.capitalize() for word in snake_case.split('_'))
+    return "".join(word.capitalize() for word in snake_case.split("_"))
 
 
 def to_snake_case(name):
@@ -127,4 +127,4 @@ def to_boto(member_vars, member_name_to_boto_name, member_name_to_type):
         else:
             boto_value = api_type.to_boto(member_value) if api_type else member_value
         to_boto_values[boto_name] = boto_value
-    return to_boto_values
+    return to_boto_values

sagemaker-core/src/sagemaker/core/base_deserializers.py

@@ -31,5 +31,5 @@
     "Importing from sagemaker.core.base_deserializers is deprecated. "
     "Use sagemaker.core.deserializers instead.",
     DeprecationWarning,
-    stacklevel=2
+    stacklevel=2,
 )

sagemaker-core/src/sagemaker/core/base_serializers.py

@@ -31,5 +31,5 @@
     "Importing from sagemaker.core.base_serializers is deprecated. "
     "Use sagemaker.core.serializers instead.",
     DeprecationWarning,
-    stacklevel=2
+    stacklevel=2,
 )

sagemaker-core/src/sagemaker/core/clarify/__init__.py

@@ -2004,7 +2004,7 @@ def _run(
                 kms_key,
             )
             from sagemaker.core.shapes import ProcessingS3Input, ProcessingS3Output
-            
+
             config_input = ProcessingInput(
                 input_name="analysis_config",
                 s3_input=ProcessingS3Input(
@@ -2013,7 +2013,7 @@ def _run(
                     s3_data_type="S3Prefix",
                     s3_input_mode="File",
                     s3_compression_type="None",
-                )
+                ),
             )
             data_input = ProcessingInput(
                 input_name="dataset",
@@ -2024,15 +2024,15 @@ def _run(
                     s3_input_mode="File",
                     s3_data_distribution_type=data_config.s3_data_distribution_type,
                     s3_compression_type=data_config.s3_compression_type,
-                )
+                ),
             )
             result_output = ProcessingOutput(
                 output_name="analysis_result",
                 s3_output=ProcessingS3Output(
                     s3_uri=data_config.s3_output_path,
                     local_path=self._CLARIFY_OUTPUT,
                     s3_upload_mode=ProcessingOutputHandler.get_s3_upload_mode(analysis_config),
-                )
+                ),
             )
 
             return super().run(
@@ -2101,9 +2101,7 @@ def run_pre_training_bias(
             data_config, data_bias_config, methods
         )
         # when name is either not provided (is None) or an empty string ("")
-        job_name = job_name or name_from_base(
-            self.job_name_prefix or "Clarify-Pretraining-Bias"
-        )
+        job_name = job_name or name_from_base(self.job_name_prefix or "Clarify-Pretraining-Bias")
         return self._run(
             data_config,
             analysis_config,
@@ -2187,9 +2185,7 @@ def run_post_training_bias(
             model_config,
         )
         # when name is either not provided (is None) or an empty string ("")
-        job_name = job_name or name_from_base(
-            self.job_name_prefix or "Clarify-Posttraining-Bias"
-        )
+        job_name = job_name or name_from_base(self.job_name_prefix or "Clarify-Posttraining-Bias")
         return self._run(
             data_config,
             analysis_config,
@@ -2373,9 +2369,7 @@ def run_explainability(
             data_config, model_config, model_scores, explainability_config
         )
         # when name is either not provided (is None) or an empty string ("")
-        job_name = job_name or name_from_base(
-            self.job_name_prefix or "Clarify-Explainability"
-        )
+        job_name = job_name or name_from_base(self.job_name_prefix or "Clarify-Explainability")
         return self._run(
             data_config,
             analysis_config,

sagemaker-core/src/sagemaker/core/common_utils.py

@@ -88,6 +88,7 @@ class ModelApprovalStatusEnum(str, Enum):
     REJECTED = "Rejected"
     PENDING_MANUAL_APPROVAL = "PendingManualApproval"
 
+
 # Use the base name of the image as the job name if the user doesn't give us one
 def name_from_image(image, max_length=63):
     """Create a training job name based on the image name and a timestamp.
@@ -1136,6 +1137,7 @@ def resolve_value_from_config(
         else None
     )
     from sagemaker.core.config.config_utils import _log_sagemaker_config_single_substitution
+
     _log_sagemaker_config_single_substitution(direct_input, config_value, config_path)
 
     if direct_input is not None:
@@ -1179,6 +1181,7 @@ def get_sagemaker_config_value(sagemaker_session, key, sagemaker_config: dict =
     # Copy the value so any modifications to the output will not modify the source config
     return copy.deepcopy(config_value)
 
+
 def get_resource_name_from_arn(arn):
     """Extract the resource name from an ARN string.
 
@@ -1190,6 +1193,7 @@ def get_resource_name_from_arn(arn):
     """
     return arn.split(":", 5)[5].split("/", 1)[1]
 
+
 def list_tags(sagemaker_session, resource_arn, max_results=50):
     """List the tags given an Amazon Resource Name.
 
@@ -1223,6 +1227,7 @@ def list_tags(sagemaker_session, resource_arn, max_results=50):
         logger.error("Error retrieving tags. resource_arn: %s", resource_arn)
         raise error
 
+
 def resolve_class_attribute_from_config(
     clazz: Optional[type],
     instance: Optional[object],
@@ -1290,6 +1295,7 @@ def resolve_class_attribute_from_config(
             setattr(instance, attribute, default_value)
 
     from sagemaker.core.config.config_utils import _log_sagemaker_config_single_substitution
+
     _log_sagemaker_config_single_substitution(current_value, config_value, config_path)
 
     return instance
@@ -1344,6 +1350,7 @@ def resolve_nested_dict_value_from_config(
             dictionary = set_nested_value(dictionary, nested_keys, default_value)
 
     from sagemaker.core.config.config_utils import _log_sagemaker_config_single_substitution
+
     _log_sagemaker_config_single_substitution(current_nested_value, config_value, config_path)
 
     return dictionary
@@ -1410,6 +1417,7 @@ def update_list_of_dicts_with_values_from_config(
         input_list[i] = dict_from_config
 
     from sagemaker.core.config.config_utils import _log_sagemaker_config_merge
+
     _log_sagemaker_config_merge(
         source_value=inputs_copy,
         config_value=unmodified_inputs_from_config,
@@ -1477,6 +1485,7 @@ def update_nested_dictionary_with_values_from_config(
         return source_dict
 
     from sagemaker.core.config.config_utils import _log_sagemaker_config_merge
+
     _log_sagemaker_config_merge(
         source_value=source_dict,
         config_value=original_config_dict_value,
@@ -2023,6 +2032,7 @@ def _walk_and_apply_json(json_obj, new):
 
     return _walk_and_apply_json(json_obj, new={})
 
+
 def _wait_until(callable_fn, poll=5):
     """Placeholder docstring"""
     elapsed_time = 0
@@ -2048,6 +2058,7 @@ def _wait_until(callable_fn, poll=5):
             raise err
     return result
 
+
 def _flush_log_streams(
     stream_names, instance_count, client, log_group, job_name, positions, dot, color_wrap
 ):
@@ -2098,7 +2109,9 @@ def _flush_log_streams(
             color_wrap(idx, event["message"])
             ts, count = positions[stream_names[idx]]
             if event["timestamp"] == ts:
-                positions[stream_names[idx]] = sagemaker.core.logs.Position(timestamp=ts, skip=count + 1)
+                positions[stream_names[idx]] = sagemaker.core.logs.Position(
+                    timestamp=ts, skip=count + 1
+                )
             else:
                 positions[stream_names[idx]] = sagemaker.core.logs.Position(
                     timestamp=event["timestamp"], skip=1
@@ -2108,6 +2121,7 @@ def _flush_log_streams(
         print(".", end="")
         sys.stdout.flush()
 
+
 class LogState(object):
     """Placeholder docstring"""
 
@@ -2117,6 +2131,7 @@ class LogState(object):
     JOB_COMPLETE = 4
     COMPLETE = 5
 
+
 _STATUS_CODE_TABLE = {
     "COMPLETED": "Completed",
     "INPROGRESS": "InProgress",
@@ -2128,12 +2143,14 @@ class LogState(object):
     "PENDING": "Pending",
 }
 
+
 def _get_initial_job_state(description, status_key, wait):
     """Placeholder docstring"""
     status = description[status_key]
     job_already_completed = status in ("Completed", "Failed", "Stopped")
     return LogState.TAILING if wait and not job_already_completed else LogState.COMPLETE
 
+
 def _logs_init(boto_session, description, job):
     """Placeholder docstring"""
     if job == "Training":
@@ -2165,6 +2182,7 @@ def _logs_init(boto_session, description, job):
 
     return instance_count, stream_names, positions, client, log_group, dot, color_wrap
 
+
 def _check_job_status(job, desc, status_key_name):
     """Check to see if the job completed successfully.
 
@@ -2218,6 +2236,7 @@ def _check_job_status(job, desc, status_key_name):
             actual_status=status,
         )
 
+
 def _create_resource(create_fn):
     """Call create function and accepts/pass when resource already exists.
 
@@ -2259,4 +2278,4 @@ def _is_s3_uri(s3_uri: Optional[str]) -> bool:
     if s3_uri is None:
         return False
 
-    return re.match("^s3://([^/]+)/?(.*)$", s3_uri) is not None    
+    return re.match("^s3://([^/]+)/?(.*)$", s3_uri) is not None

sagemaker-core/src/sagemaker/core/config/__init__.py

@@ -178,4 +178,4 @@
     LOCAL,
     LOCAL_CODE,
     CONTAINER_CONFIG,
-)
+)

sagemaker-core/src/sagemaker/core/config/config.py

@@ -28,7 +28,10 @@
 from botocore.utils import merge_dicts
 from six.moves.urllib.parse import urlparse
 from sagemaker.core.config.config_schema import SAGEMAKER_PYTHON_SDK_CONFIG_SCHEMA
-from sagemaker.core.config.config_utils import non_repeating_log_factory, get_sagemaker_config_logger
+from sagemaker.core.config.config_utils import (
+    non_repeating_log_factory,
+    get_sagemaker_config_logger,
+)
 
 logger = get_sagemaker_config_logger()
 log_info_function = non_repeating_log_factory(logger, "info")

sagemaker-core/src/sagemaker/core/config/config_schema.py

@@ -733,7 +733,6 @@ def _simple_path(*args: str):
                                     },
                                 },
                                 MODEL_TRAINER: {TYPE: OBJECT, ADDITIONAL_PROPERTIES: True},
-
                                 REMOTE_FUNCTION: {
                                     TYPE: OBJECT,
                                     ADDITIONAL_PROPERTIES: False,
@@ -1218,4 +1217,4 @@ def _simple_path(*args: str):
         },
     },
     "required": [LOCAL],
-}
+}