Raghu/baseimage update

assets/training/model_management/components/convert_model_to_mlflow/spec.yaml

@@ -13,6 +13,7 @@ environment: azureml://registries/azureml/environments/model-management/versions
 
 code: ../../src/
 command: |
+  # pip install transformers==4.48.0 torch==2.2.2 numpy==1.23.5 pandas==1.5.3 urllib3==1.26.19 --no-cache-dir --force-reinstall
   # TODO: This has one disadvantage as shell logs wont be pushed to appinsights
   set -ex
   IFS=',' read -ra pip_pkgs <<< "$[[${{inputs.extra_pip_requirements}}]]"

assets/training/model_management/components/validation_trigger_import/spec.yaml

@@ -181,7 +181,7 @@ outputs:
 
 is_deterministic: True
 
-environment: azureml://registries/azureml/environments/python-sdk-v2/versions/25
+environment: azureml://registries/azureml/environments/python-sdk-v2/versions/29
 code: ../../src
 command: python run_model_validate.py --validation-info ${{outputs.validation_info}}
 

assets/training/model_management/environments/model-management/context/Dockerfile

@@ -26,6 +26,7 @@ RUN pip install -r requirements.txt --no-cache-dir
 
 # Vulnerability fix
 RUN pip install Pillow gunicorn onnx==1.17.0 idna tqdm requests==2.32.1 tornado==6.4.2 certifi==2024.07.04 urllib3==1.26.19 scikit-learn==1.5.1 mlflow==2.20.3 mlflow-skinny==2.20.3 marshmallow==3.23.2
+transformers==4.48.0 torch==2.2.2 numpy==1.23.5 pandas==1.5.3 azureml-evaluate-mlflow --no-cache-dir --force-reinstall
 
 # List pip packages
 RUN pip list

assets/training/model_management/src/azureml/model/mgmt/processors/pyfunc/convertors.py

@@ -45,7 +45,7 @@
     MLflowSchemaLiterals as VirchowMLFlowSchemaLiterals, MLflowLiterals as VirchowMLflowLiterals
 from azureml.model.mgmt.processors.pyfunc.hibou_b.config import \
     MLflowSchemaLiterals as HibouBMLFlowSchemaLiterals, MLflowLiterals as HibouBMLflowLiterals
-
+from azureml.model.mgmt.utils.common_utils import get_mlclient
 
 logger = get_logger(__name__)
 
@@ -114,6 +114,12 @@ def _save(
         :type metadata: Optional[Dict]. Defaults to {}.
         """
         signatures = self._signatures or self.get_model_signature()
+        if not self._vllm_enabled:
+            mlclient = get_mlclient("azureml")
+            mlFlow_image = mlclient.environments.get("mlflow-model-inference", label="latest")
+            metadata["azureml.base_image"] = "mcr.microsoft.com/azureml/curated/mlflow-model-inference:" \
+                + str(mlFlow_image.version)
+            logger.info("Metadata: {}".format(metadata))
         # set metadata info
         metadata.update(fetch_mlflow_acft_metadata(
             base_model_name=self._model_id,

assets/training/model_management/src/azureml/model/mgmt/processors/transformers/convertors.py

@@ -183,6 +183,12 @@ def _save(
             metadata["azureml.base_image"] = "mcr.microsoft.com/azureml/curated/foundation-model-inference:" \
                 + str(vllm_image.version)
             logger.info("Metadata: {}".format(metadata))
+        else:
+            mlclient = get_mlclient("azureml")
+            mlFlow_image = mlclient.environments.get("mlflow-model-inference", label="latest")
+            metadata["azureml.base_image"] = "mcr.microsoft.com/azureml/curated/mlflow-model-inference:" \
+                + str(mlFlow_image.version)
+            logger.info("Metadata: {}".format(metadata))
 
         if self._model_flavor == "OSS":
             try: