sagemaker 2.254.1__tar.gz → 2.254.2.dev0__tar.gz

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: sagemaker
-Version: 2.254.1
+Version: 2.254.2.dev0
 Summary: Open source library for training and deploying models on Amazon SageMaker.
 Project-URL: Homepage, https://github.com/aws/sagemaker-python-sdk
 Author: Amazon Web Services
@@ -16,7 +16,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Python: >=3.9
 Requires-Dist: attrs<26,>=24
-Requires-Dist: boto3<2.0,>=1.39.5
+Requires-Dist: boto3<2.0,>=1.42.2
 Requires-Dist: cloudpickle>=2.2.1
 Requires-Dist: docker
 Requires-Dist: fastapi
@@ -34,7 +34,7 @@ Requires-Dist: protobuf<6.32,>=3.12
 Requires-Dist: psutil
 Requires-Dist: pyyaml>=6.0.1
 Requires-Dist: requests
-Requires-Dist: sagemaker-core<2.0.0,>=1.0.17
+Requires-Dist: sagemaker-core<2.0.0,>=1.0.71
 Requires-Dist: schema
 Requires-Dist: smdebug-rulesconfig==1.0.1
 Requires-Dist: tblib<4,>=1.7.0
@@ -89,6 +89,7 @@ Requires-Dist: fastapi==0.115.4; extra == 'test'
 Requires-Dist: fastapi>=0.111.0; extra == 'test'
 Requires-Dist: flake8==7.1.2; extra == 'test'
 Requires-Dist: flask-limiter==3.11; extra == 'test'
+Requires-Dist: google-re2<1.1.20250805; (python_version < '3.10') and extra == 'test'
 Requires-Dist: huggingface-hub==0.26.2; extra == 'test'
 Requires-Dist: jinja2==3.1.6; extra == 'test'
 Requires-Dist: jsonpickle<4.0.0; extra == 'test'

sagemaker-2.254.2.dev0/VERSION

@@ -0,0 +1 @@
+2.254.2.dev0

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/pyproject.toml

@@ -32,7 +32,7 @@ classifiers = [
 ]
 dependencies = [
   "attrs>=24,<26",
-  "boto3>=1.39.5,<2.0",
+  "boto3>=1.42.2,<2.0",
   "cloudpickle>=2.2.1",
   "docker",
   "fastapi",
@@ -49,7 +49,7 @@ dependencies = [
   "psutil",
   "PyYAML>=6.0.1",
   "requests",
-  "sagemaker-core>=1.0.17,<2.0.0",
+  "sagemaker-core>=1.0.71,<2.0.0",
   "schema",
   "smdebug_rulesconfig==1.0.1",
   "tblib>=1.7.0,<4",

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/requirements/extras/test_requirements.txt

@@ -61,3 +61,5 @@ mypy-boto3-redshift-data==1.35.51
 mypy-boto3-s3==1.35.76
 mypy-extensions==1.0.0
 mypy==1.9.0
+# apache-airflow transitive dependancy
+google-re2<1.1.20250805; python_version < "3.10"

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/__init__.py

@@ -48,6 +48,7 @@ from sagemaker.algorithm import AlgorithmEstimator  # noqa: F401
 from sagemaker.analytics import TrainingJobAnalytics, HyperparameterTuningJobAnalytics  # noqa: F401
 from sagemaker.local.local_session import LocalSession  # noqa: F401
 
+from sagemaker.container_base_model import ContainerBaseModel  # noqa: F401
 from sagemaker.model import Model, ModelPackage  # noqa: F401
 from sagemaker.model_metrics import ModelMetrics, MetricsSource, FileSource  # noqa: F401
 from sagemaker.pipeline import PipelineModel  # noqa: F401

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/chainer/model.py

@@ -17,7 +17,7 @@ import logging
 from typing import Callable, Optional, Union, List, Dict
 
 import sagemaker
-from sagemaker import image_uris, ModelMetrics
+from sagemaker import image_uris, ModelMetrics, ContainerBaseModel
 from sagemaker.drift_check_baselines import DriftCheckBaselines
 from sagemaker.fw_utils import (
     model_code_key_prefix,
@@ -182,6 +182,8 @@ class ChainerModel(FrameworkModel):
         source_uri: Optional[Union[str, PipelineVariable]] = None,
         model_card: Optional[Union[ModelPackageModelCard, ModelCard]] = None,
         model_life_cycle: Optional[ModelLifeCycle] = None,
+        model_package_registration_type: Optional[Union[str, PipelineVariable]] = None,
+        base_model: Optional[ContainerBaseModel] = None,
     ):
         """Creates a model package for creating SageMaker models or listing on Marketplace.
 
@@ -236,6 +238,9 @@ class ChainerModel(FrameworkModel):
             model_card (ModeCard or ModelPackageModelCard): document contains qualitative and
                 quantitative information about a model (default: None).
             model_life_cycle (ModelLifeCycle): ModelLifeCycle object (default: None).
+            model_package_registration_type (str or PipelineVariable): Model Package Registration
+                Type (default: None).
+            base_model (ContainerBaseModel): ContainerBaseModel object (default: None).
 
         Returns:
             str: A string of SageMaker Model Package ARN.
@@ -278,6 +283,8 @@ class ChainerModel(FrameworkModel):
             source_uri=source_uri,
             model_card=model_card,
             model_life_cycle=model_life_cycle,
+            model_package_registration_type=model_package_registration_type,
+            base_model=base_model,
         )
 
     def prepare_container_def(

sagemaker-2.254.2.dev0/src/sagemaker/container_base_model.py

@@ -0,0 +1,52 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express oXr implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""This file contains code related to base model for containers."""
+from __future__ import absolute_import
+
+from typing import Optional, Union
+
+from sagemaker.workflow.entities import PipelineVariable
+
+
+class ContainerBaseModel(object):
+    """Accepts Base Model parameters for conversion to request dict."""
+
+    def __init__(
+        self,
+        hub_content_name: Union[str, PipelineVariable] = None,
+        hub_content_version: Optional[Union[str, PipelineVariable]] = None,
+        recipe_name: Optional[Union[str, PipelineVariable]] = None,
+    ):
+        """Initialize a ``ContainerBaseModel`` instance and turn parameters into dict.
+
+        Args:
+            hub_content_name (str or PipelineVariable): The hub content name
+            hub_content_version (str or PipelineVariable): The hub content version
+                (default: None)
+            recipe_name (str or PipelineVariable): The Recipe name
+               (default: None)
+        """
+        self.hub_content_name = hub_content_name
+        self.hub_content_version = hub_content_version
+        self.recipe_name = recipe_name
+
+    def _to_request_dict(self):
+        """Generates a request dictionary using the parameters provided to the class."""
+        base_model_request = {}
+        if self.hub_content_name is not None:
+            base_model_request["HubContentName"] = self.hub_content_name
+        if self.hub_content_version is not None:
+            base_model_request["HubContentVersion"] = self.hub_content_version
+        if self.recipe_name is not None:
+            base_model_request["RecipeName"] = self.recipe_name
+        return base_model_request

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/estimator.py

@@ -1817,6 +1817,8 @@ class EstimatorBase(with_metaclass(ABCMeta, object)):  # pylint: disable=too-man
         source_uri=None,
         model_life_cycle=None,
         model_card=None,
+        model_package_registration_type=None,
+        base_model=None,
         **kwargs,
     ):
         """Creates a model package for creating SageMaker models or listing on Marketplace.
@@ -1868,6 +1870,9 @@ class EstimatorBase(with_metaclass(ABCMeta, object)):  # pylint: disable=too-man
             model_card (ModeCard or ModelPackageModelCard): document contains qualitative and
                 quantitative information about a model (default: None).
             model_life_cycle (ModelLifeCycle): ModelLifeCycle object (default: None).
+            model_package_registration_type (str): Model Package Registration
+                Type (default: None).
+            base_model (ContainerBaseModel): ContainerBaseModel object (default: None).
             **kwargs: Passed to invocation of ``create_model()``. Implementations may customize
                 ``create_model()`` to accept ``**kwargs`` to customize model creation during
                 deploy. For more, see the implementation docs.
@@ -1924,6 +1929,8 @@ class EstimatorBase(with_metaclass(ABCMeta, object)):  # pylint: disable=too-man
             source_uri=source_uri,
             model_card=model_card,
             model_life_cycle=model_life_cycle,
+            model_package_registration_type=model_package_registration_type,
+            base_model=base_model,
         )
 
     @property

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/huggingface/llm_utils.py

@@ -67,6 +67,14 @@ def get_huggingface_llm_image_uri(
             image_scope="inference",
             inference_tool="neuronx",
         )
+    if backend == "huggingface-vllm-neuronx":
+        return image_uris.retrieve(
+            "huggingface-vllm-neuronx",
+            region=region,
+            version=version,
+            image_scope="inference",
+            inference_tool="neuronx",
+        )
     if backend == "huggingface-tei":
         return image_uris.retrieve(
             "huggingface-tei",

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/huggingface/model.py

@@ -17,7 +17,7 @@ import logging
 from typing import Callable, Optional, Union, List, Dict
 
 import sagemaker
-from sagemaker import image_uris, ModelMetrics
+from sagemaker import image_uris, ModelMetrics, ContainerBaseModel
 from sagemaker.deserializers import JSONDeserializer
 from sagemaker.drift_check_baselines import DriftCheckBaselines
 from sagemaker.fw_utils import (
@@ -372,6 +372,8 @@ class HuggingFaceModel(FrameworkModel):
         source_uri: Optional[Union[str, PipelineVariable]] = None,
         model_life_cycle: Optional[ModelLifeCycle] = None,
         model_card: Optional[Union[ModelPackageModelCard, ModelCard]] = None,
+        model_package_registration_type: Optional[Union[str, PipelineVariable]] = None,
+        base_model: Optional[ContainerBaseModel] = None,
     ):
         """Creates a model package for creating SageMaker models or listing on Marketplace.
 
@@ -427,6 +429,9 @@ class HuggingFaceModel(FrameworkModel):
             model_card (ModeCard or ModelPackageModelCard): document contains qualitative and
                 quantitative information about a model (default: None).
             model_life_cycle (ModelLifeCycle): ModelLifeCycle object (default: None).
+            model_package_registration_type (str or PipelineVariable): Model Package Registration
+                Type (default: None).
+            base_model (ContainerBaseModel): ContainerBaseModel object (default: None).
 
         Returns:
             A `sagemaker.model.ModelPackage` instance.
@@ -477,6 +482,8 @@ class HuggingFaceModel(FrameworkModel):
             source_uri=source_uri,
             model_life_cycle=model_life_cycle,
             model_card=model_card,
+            model_package_registration_type=model_package_registration_type,
+            base_model=base_model,
         )
 
     def prepare_container_def(

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/image_uri_config/huggingface-neuronx.json

@@ -9,7 +9,8 @@
             "4.36": "4.36.2",
             "4.43": "4.43.2",
             "4.48": "4.48.1",
-            "4.51": "4.51.0"
+            "4.51": "4.51.0",
+            "4.55": "4.55.4"
         },
         "versions": {
             "4.28.1": {
@@ -272,6 +273,49 @@
                         "sdk2.24.1"
                     ]
                 }
+            },
+            "4.55.4": {
+                "version_aliases": {
+                    "pytorch2.7": "pytorch2.7.0"
+                },
+                "pytorch2.7.0": {
+                    "py_versions": [
+                        "py310"
+                    ],
+                    "repository": "huggingface-pytorch-training-neuronx",
+                    "registries": {
+                        "ap-northeast-1": "763104351884",
+                        "ap-south-1": "763104351884",
+                        "ap-south-2": "772153158452",
+                        "ap-southeast-1": "763104351884",
+                        "ap-southeast-2": "763104351884",
+                        "ap-southeast-4": "457447274322",
+                        "ap-southeast-5": "550225433462",
+                        "ap-southeast-7": "590183813437",
+                        "cn-north-1": "727897471807",
+                        "cn-northwest-1": "727897471807",
+                        "eu-central-1": "763104351884",
+                        "eu-central-2": "380420809688",
+                        "eu-south-2": "503227376785",
+                        "eu-west-1": "763104351884",
+                        "eu-west-3": "763104351884",
+                        "il-central-1": "780543022126",
+                        "mx-central-1":"637423239942",
+                        "sa-east-1": "763104351884",
+                        "us-east-1": "763104351884",
+                        "us-east-2": "763104351884",
+                        "us-gov-east-1": "446045086412",
+                        "us-gov-west-1": "442386744353",
+                        "us-west-2": "763104351884",
+                        "ca-west-1": "204538143572"
+                    },
+                    "container_version": {
+                        "inf": "ubuntu22.04"
+                    },
+                    "sdk_versions": [
+                        "sdk2.26.0"
+                    ]
+                }
             }
         }
     },
@@ -284,7 +328,8 @@
             "4.34": "4.34.1",
             "4.36": "4.36.2",
             "4.43": "4.43.2",
-            "4.51": "4.51.3"
+            "4.51": "4.51.3",
+            "4.55": "4.55.4"
         },
         "versions": {
             "4.28.1": {
@@ -592,6 +637,49 @@
                         "sdk2.24.1"
                     ]
                 }
+            },
+            "4.55.4": {
+                "version_aliases": {
+                    "pytorch2.7": "pytorch2.7.1"
+                },
+                "pytorch2.7.1": {
+                    "py_versions": [
+                        "py310"
+                    ],
+                    "repository": "huggingface-pytorch-inference-neuronx",
+                    "registries": {
+                        "ap-northeast-1": "763104351884",
+                        "ap-south-1": "763104351884",
+                        "ap-south-2": "772153158452",
+                        "ap-southeast-1": "763104351884",
+                        "ap-southeast-2": "763104351884",
+                        "ap-southeast-4": "457447274322",
+                        "ap-southeast-5": "550225433462",
+                        "ap-southeast-7": "590183813437",
+                        "cn-north-1": "727897471807",
+                        "cn-northwest-1": "727897471807",
+                        "eu-central-1": "763104351884",
+                        "eu-central-2": "380420809688",
+                        "eu-south-2": "503227376785",
+                        "eu-west-1": "763104351884",
+                        "eu-west-3": "763104351884",
+                        "il-central-1": "780543022126",
+                        "mx-central-1":"637423239942",
+                        "sa-east-1": "763104351884",
+                        "us-east-1": "763104351884",
+                        "us-east-2": "763104351884",
+                        "us-gov-east-1": "446045086412",
+                        "us-gov-west-1": "442386744353",
+                        "us-west-2": "763104351884",
+                        "ca-west-1": "204538143572"
+                    },
+                    "container_version": {
+                        "inf": "ubuntu22.04"
+                    },
+                    "sdk_versions": [
+                        "sdk2.26.0"
+                    ]
+                }
             }
         }
     }

sagemaker-2.254.2.dev0/src/sagemaker/image_uri_config/huggingface-vllm-neuronx.json

@@ -0,0 +1,38 @@
+{
+    "inference": {
+        "processors": [
+            "inf2"
+        ],
+        "version_aliases": {
+            "0.4": "0.4.1"
+        },
+        "versions": {
+            "0.4.1": {
+                "py_versions": [
+                    "py310"
+                ],
+                "registries": {
+                    "ap-northeast-1": "763104351884",
+                    "ap-south-1": "763104351884",
+                    "ap-southeast-1": "763104351884",
+                    "ap-southeast-2": "763104351884",
+                    "eu-central-1": "763104351884",
+                    "eu-west-1": "763104351884",
+                    "eu-west-3": "763104351884",
+                    "sa-east-1": "763104351884",
+                    "us-east-1": "763104351884",
+                    "us-east-2": "763104351884",
+                    "us-west-2": "763104351884"
+                },
+                "tag_prefix": "0.10.2",
+                "repository": "huggingface-vllm-inference-neuronx",
+                "container_version": {
+                    "inf2": "ubuntu22.04"
+                },
+                "sdk_versions": [
+                    "sdk2.26.0"
+                ]
+            }
+        }
+    }
+}

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/image_uris.py

@@ -41,6 +41,7 @@ HUGGING_FACE_LLM_FRAMEWORK = "huggingface-llm"
 HUGGING_FACE_TEI_GPU_FRAMEWORK = "huggingface-tei"
 HUGGING_FACE_TEI_CPU_FRAMEWORK = "huggingface-tei-cpu"
 HUGGING_FACE_LLM_NEURONX_FRAMEWORK = "huggingface-llm-neuronx"
+HUGGING_FACE_VLLM_NEURONX_FRAMEWORK = "huggingface-vllm-neuronx"
 XGBOOST_FRAMEWORK = "xgboost"
 SKLEARN_FRAMEWORK = "sklearn"
 TRAINIUM_ALLOWED_FRAMEWORKS = "pytorch"
@@ -230,7 +231,11 @@ def retrieve(
         container_version = version_config["container_version"][processor]
 
     # Append sdk version in case of trainium instances
-    if repo in ["pytorch-training-neuron", "pytorch-training-neuronx"]:
+    if repo in [
+        "pytorch-training-neuron",
+        "pytorch-training-neuronx",
+        "huggingface-vllm-inference-neuronx",
+    ]:
         if not sdk_version:
             sdk_version = _get_latest_versions(version_config["sdk_versions"])
         container_version = sdk_version + "-" + container_version

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/model.py

@@ -52,6 +52,7 @@ from sagemaker.model_card import (
 from sagemaker.model_card.helpers import _hash_content_str
 from sagemaker.model_card.schema_constraints import ModelApprovalStatusEnum
 from sagemaker.session import Session
+from sagemaker.container_base_model import ContainerBaseModel
 from sagemaker.model_metrics import ModelMetrics
 from sagemaker.drift_check_baselines import DriftCheckBaselines
 from sagemaker.explainer import ExplainerConfig
@@ -477,6 +478,8 @@ class Model(ModelBase, InferenceRecommenderMixin):
         model_life_cycle: Optional[ModelLifeCycle] = None,
         accept_eula: Optional[bool] = None,
         model_type: Optional[JumpStartModelType] = None,
+        model_package_registration_type: Optional[Union[str, PipelineVariable]] = None,
+        base_model: Optional[ContainerBaseModel] = None,
     ):
         """Creates a model package for creating SageMaker models or listing on Marketplace.
 
@@ -531,6 +534,9 @@ class Model(ModelBase, InferenceRecommenderMixin):
             model_card (ModeCard or ModelPackageModelCard): document contains qualitative and
                 quantitative information about a model (default: None).
             model_life_cycle (ModelLifeCycle): ModelLifeCycle object (default: None).
+            model_package_registration_type (str or PipelineVariable): Model Package Registration
+                Type (default: None).
+            base_model (ContainerBaseModel): ContainerBaseModel object (default: None).
 
         Returns:
             A `sagemaker.model.ModelPackage` instance or pipeline step arguments
@@ -578,6 +584,9 @@ class Model(ModelBase, InferenceRecommenderMixin):
             if self.model_data is not None:
                 container_def["ModelDataUrl"] = self.model_data
 
+        if base_model is not None and hasattr(base_model, "_to_request_dict"):
+            container_def["BaseModel"] = base_model._to_request_dict()
+
         model_pkg_args = sagemaker.get_model_package_args(
             self.content_types,
             self.response_types,
@@ -601,6 +610,8 @@ class Model(ModelBase, InferenceRecommenderMixin):
             source_uri=source_uri,
             model_card=model_card,
             model_life_cycle=model_life_cycle,
+            model_package_registration_type=model_package_registration_type,
+            base_model=base_model,
         )
         model_package = self.sagemaker_session.create_model_package_from_containers(
             **model_pkg_args
@@ -2150,6 +2161,7 @@ class FrameworkModel(Model):
             You can find additional parameters for initializing this class at
             :class:`~sagemaker.model.Model`.
         """
+
         super(FrameworkModel, self).__init__(
             image_uri,
             model_data,

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/modules/train/model_trainer.py

@@ -108,6 +108,7 @@ from sagemaker.modules.train.sm_recipes.utils import (
     _get_args_from_recipe,
     _determine_device_type,
     _is_nova_recipe,
+    _is_llmft_recipe,
     _load_base_recipe,
 )
 
@@ -252,6 +253,7 @@ class ModelTrainer(BaseModel):
     _metric_definitions: Optional[List[MetricDefinition]] = PrivateAttr(default=None)
 
     _is_nova_recipe: Optional[bool] = PrivateAttr(default=None)
+    _is_llmft_recipe: Optional[bool] = PrivateAttr(default=None)
     _temp_recipe_train_dir: Optional[TemporaryDirectory] = PrivateAttr(default=None)
     _temp_code_dir: Optional[TemporaryDirectory] = PrivateAttr(default=None)
 
@@ -632,12 +634,13 @@ class ModelTrainer(BaseModel):
 
             final_input_data_config = list(existing_channels.values()) + new_channels
 
-        if self._is_nova_recipe:
+        if self._is_nova_recipe or self._is_llmft_recipe:
+
             for input_data in final_input_data_config:
                 if input_data.channel_name == SM_RECIPE:
                     raise ValueError(
                         "Cannot use reserved channel name 'recipe' as an input channel name "
-                        " for Nova Recipe"
+                        " for Nova or LLMFT Recipe"
                     )
             recipe_file_path = os.path.join(self._temp_recipe_train_dir.name, SM_RECIPE_YAML)
             recipe_channel = self.create_input_data_channel(
@@ -646,7 +649,10 @@ class ModelTrainer(BaseModel):
                 key_prefix=input_data_key_prefix,
             )
             final_input_data_config.append(recipe_channel)
-            self.hyperparameters.update({"sagemaker_recipe_local_path": SM_RECIPE_CONTAINER_PATH})
+            if self._is_nova_recipe:
+                self.hyperparameters.update(
+                    {"sagemaker_recipe_local_path": SM_RECIPE_CONTAINER_PATH}
+                )
 
         if final_input_data_config:
             final_input_data_config = self._get_input_data_config(
@@ -1201,14 +1207,15 @@ class ModelTrainer(BaseModel):
             training_recipe=training_recipe, recipe_overrides=recipe_overrides
         )
         is_nova = _is_nova_recipe(recipe=recipe)
+        is_llmft = _is_llmft_recipe(recipe=recipe)
 
-        if device_type == "cpu" and not is_nova:
+        if device_type == "cpu" and not (is_nova or is_llmft):
             raise ValueError(
                 "Training recipe is not supported for CPU instances. "
                 + "Please provide a GPU or Tranium instance type."
             )
-        if training_image is None and is_nova:
-            raise ValueError("training_image must be provided when using recipe for Nova.")
+        if training_image is None and (is_nova or is_llmft):
+            raise ValueError("training_image must be provided when using recipe for Nova or LLMFT")
 
         if training_image_config and training_image is None:
             raise ValueError("training_image must be provided when using training_image_config.")
@@ -1238,7 +1245,7 @@ class ModelTrainer(BaseModel):
             model_trainer_args["training_image"] = training_image
         if hyperparameters and not is_nova:
             logger.warning(
-                "Hyperparameters are not supported for general training recipes. "
+                "Hyperparameters are not supported for general and LLMFT training recipes. "
                 + "Ignoring hyperparameters input."
             )
         if is_nova:
@@ -1264,6 +1271,7 @@ class ModelTrainer(BaseModel):
             **model_trainer_args,
         )
         model_trainer._is_nova_recipe = is_nova
+        model_trainer._is_llmft_recipe = is_llmft
         model_trainer._temp_recipe_train_dir = tmp_dir
         return model_trainer
 

{sagemaker-2.254.1 → sagemaker-2.254.2.dev0}/src/sagemaker/modules/train/sm_recipes/utils.py

@@ -46,6 +46,19 @@ def _try_resolve_recipe(recipe: DictConfig, key=None) -> DictConfig:
     return recipe[key]
 
 
+def _resolve_final_recipe(recipe: DictConfig):
+    """Resolve final recipe."""
+    final_recipe = _try_resolve_recipe(recipe)
+    if final_recipe is None:
+        final_recipe = _try_resolve_recipe(recipe, "recipes")
+    if final_recipe is None:
+        final_recipe = _try_resolve_recipe(recipe, "training")
+    if final_recipe is None:
+        raise RuntimeError("Could not resolve provided recipe.")
+
+    return final_recipe
+
+
 def _determine_device_type(instance_type: str) -> str:
     """Determine device type (gpu, cpu, trainium) based on instance type."""
     instance_family = instance_type.split(".")[1]
@@ -269,6 +282,27 @@ def _is_nova_recipe(
     return bool(has_nova_model) or bool(has_distillation)
 
 
+def _is_llmft_recipe(
+    recipe: DictConfig,
+) -> bool:
+    """Check if the recipe is a LLMFT recipe.
+
+    A recipe is considered a LLMFT recipe if it meets the following conditions:
+        1. Having a run section
+        2. The model_type in run is llmft
+        3. Having a training_config section
+
+    Args:
+        recipe (DictConfig): The loaded recipe configuration
+
+    Returns:
+        bool: True if the recipe is a LLMFT recipe, False otherwise
+    """
+    run_config = recipe.get("run", {})
+    has_llmft_model = run_config.get("model_type", "").lower() == "llm_finetuning_aws"
+    return bool(has_llmft_model) and bool(recipe.get("training_config"))
+
+
 def _get_args_from_nova_recipe(
     recipe: DictConfig,
     compute: Compute,
@@ -312,6 +346,12 @@ def _get_args_from_nova_recipe(
         if lambda_arn:
             args["hyperparameters"]["eval_lambda_arn"] = lambda_arn
 
+    # Handle reward lambda configuration
+    run_config = recipe.get("run", {})
+    reward_lambda_arn = run_config.get("reward_lambda_arn", "")
+    if reward_lambda_arn:
+        args["hyperparameters"]["reward_lambda_arn"] = reward_lambda_arn
+
     _register_custom_resolvers()
 
     # Resolve Final Recipe
@@ -339,6 +379,43 @@ def _get_args_from_nova_recipe(
     return args, recipe_local_dir
 
 
+def _get_args_from_llmft_recipe(
+    recipe: DictConfig,
+    compute: Compute,
+) -> Tuple[Dict[str, Any], tempfile.TemporaryDirectory]:
+
+    if not compute.instance_count and not recipe.get("trainer", {}).get("num_nodes", None):
+        raise ValueError(
+            "Must set ``instance_count`` in compute or ``num_nodes`` in trainer in recipe."
+        )
+    if compute.instance_count and recipe.get("trainer", {}).get("num_nodes", None) is not None:
+        logger.warning(
+            f"Using Compute to set instance_count:\n{compute}."
+            "\nIgnoring trainer -> num_nodes in recipe."
+        )
+    compute.instance_count = compute.instance_count or recipe.get("trainer", {}).get("num_nodes")
+
+    args = dict()
+
+    _register_custom_resolvers()
+    final_recipe = _resolve_final_recipe(recipe)
+
+    # Save Final Recipe to tmp dir
+    recipe_local_dir = tempfile.TemporaryDirectory(prefix="recipe_")
+    final_recipe_path = os.path.join(recipe_local_dir.name, SM_RECIPE_YAML)
+    OmegaConf.save(config=final_recipe, f=final_recipe_path)
+
+    args.update(
+        {
+            "compute": compute,
+            "training_image": None,
+            "source_code": None,
+            "distributed": None,
+        }
+    )
+    return args, recipe_local_dir
+
+
 def _get_args_from_recipe(
     training_recipe: Union[str, DictConfig],
     compute: Compute,
@@ -383,6 +460,9 @@ def _get_args_from_recipe(
     if _is_nova_recipe(recipe):
         args, recipe_local_dir = _get_args_from_nova_recipe(recipe, compute, role=role)
         return args, recipe_local_dir
+    if _is_llmft_recipe(recipe):
+        args, recipe_local_dir = _get_args_from_llmft_recipe(recipe, compute)
+        return args, recipe_local_dir
 
     if "trainer" not in recipe:
         raise ValueError("Supplied recipe does not contain required field trainer.")