Added cli to convert legacy fine tuned model to v2. (#1241)

lu-ohai · web-flow · commit d852209ad9b1 · 2025-08-06T10:48:41.000-07:00
diff --git a/ads/aqua/constants.py b/ads/aqua/constants.py
@@ -45,6 +45,7 @@
 AQUA_TROUBLESHOOTING_LINK = "https://github.com/oracle-samples/oci-data-science-ai-samples/blob/main/ai-quick-actions/troubleshooting-tips.md"
 MODEL_FILE_DESCRIPTION_VERSION = "1.0"
 MODEL_FILE_DESCRIPTION_TYPE = "modelOSSReferenceDescription"
+AQUA_FINE_TUNE_MODEL_VERSION = "v2"
 
 TRAINING_METRICS_FINAL = "training_metrics_final"
 VALIDATION_METRICS_FINAL = "validation_metrics_final"
diff --git a/ads/aqua/model/model.py b/ads/aqua/model/model.py
@@ -43,6 +43,7 @@
 )
 from ads.aqua.config.container_config import AquaContainerConfig
 from ads.aqua.constants import (
+    AQUA_FINE_TUNE_MODEL_VERSION,
     AQUA_MODEL_ARTIFACT_CONFIG,
     AQUA_MODEL_ARTIFACT_CONFIG_MODEL_NAME,
     AQUA_MODEL_ARTIFACT_CONFIG_MODEL_TYPE,
@@ -645,6 +646,89 @@ def edit_registered_model(
         else:
             raise AquaRuntimeError("Only registered unverified models can be edited.")
 
+    def convert_fine_tune(
+        self, model_id: str, delete_model: Optional[bool] = False
+    ) -> DataScienceModel:
+        """Converts legacy fine tuned model to fine tuned model v2.
+        1. 'fine_tune_model_version' tag will be added as 'v2' to new fine tuned model.
+        2. 'model_file_description' json will only contain fine tuned artifacts for new fine tuned model.
+
+        Parameters
+        ----------
+        model_id: str
+            The legacy fine tuned model OCID.
+        delete_model: bool
+            Flag whether to delete the legacy model or not. Defaults to False.
+
+        Returns
+        -------
+        DataScienceModel:
+            The instance of DataScienceModel.
+        """
+        legacy_fine_tuned_model = DataScienceModel.from_id(model_id)
+        legacy_tags = legacy_fine_tuned_model.freeform_tags or {}
+
+        if (
+            Tags.AQUA_TAG not in legacy_tags
+            or Tags.AQUA_FINE_TUNED_MODEL_TAG not in legacy_tags
+        ):
+            raise AquaValueError(
+                f"Model '{model_id}' is not eligible for conversion. Only legacy AQUA fine-tuned models "
+                f"without the 'fine_tune_model_version={AQUA_FINE_TUNE_MODEL_VERSION}' tag are supported."
+            )
+
+        if (
+            legacy_tags.get(Tags.AQUA_FINE_TUNE_MODEL_VERSION, UNKNOWN).lower()
+            == AQUA_FINE_TUNE_MODEL_VERSION
+        ):
+            raise AquaValueError(
+                f"Model '{model_id}' is already a fine-tuned model in version '{AQUA_FINE_TUNE_MODEL_VERSION}'. "
+                "No conversion is necessary."
+            )
+
+        if not legacy_fine_tuned_model.model_file_description:
+            raise AquaValueError(
+                f"Model '{model_id}' is missing required metadata and cannot be converted. "
+                "This may indicate the model was not created properly or is not a supported legacy AQUA fine-tuned model."
+            )
+
+        # add 'fine_tune_model_version' tag as 'v2'
+        fine_tune_model_v2_tags = {
+            **legacy_tags,
+            Tags.AQUA_FINE_TUNE_MODEL_VERSION: AQUA_FINE_TUNE_MODEL_VERSION,
+        }
+
+        # remove base model artifacts in 'model_file_description' json file
+        # base model artifacts are placed as the first entry in 'models' list
+        legacy_fine_tuned_model.model_file_description["models"].pop(0)
+
+        fine_tune_model_v2 = (
+            DataScienceModel()
+            .with_compartment_id(legacy_fine_tuned_model.compartment_id)
+            .with_project_id(legacy_fine_tuned_model.project_id)
+            .with_model_file_description(
+                json_dict=legacy_fine_tuned_model.model_file_description
+            )
+            .with_display_name(legacy_fine_tuned_model.display_name)
+            .with_description(legacy_fine_tuned_model.description)
+            .with_freeform_tags(**fine_tune_model_v2_tags)
+            .with_defined_tags(**(legacy_fine_tuned_model.defined_tags or {}))
+            .with_custom_metadata_list(legacy_fine_tuned_model.custom_metadata_list)
+            .with_defined_metadata_list(legacy_fine_tuned_model.defined_metadata_list)
+            .with_provenance_metadata(legacy_fine_tuned_model.provenance_metadata)
+            .create(model_by_reference=True)
+        )
+
+        logger.info(
+            f"Successfully created version '{AQUA_FINE_TUNE_MODEL_VERSION}' fine-tuned model: '{fine_tune_model_v2.id}' "
+            f"based on legacy model '{model_id}'. This new model is now ready for deployment."
+        )
+
+        if delete_model:
+            legacy_fine_tuned_model.delete()
+
+        return fine_tune_model_v2
+
     def _fetch_metric_from_metadata(
         self,
         custom_metadata_list: ModelCustomMetadata,
diff --git a/tests/unitary/with_extras/aqua/test_model.py b/tests/unitary/with_extras/aqua/test_model.py
@@ -825,6 +825,125 @@ def test_get_model_fine_tuned(
             "evaluation_container": "odsc-llm-evaluate",
         }
 
+    @patch.object(DataScienceModel, "create")
+    @patch.object(DataScienceModel, "from_id")
+    def test_convert_fine_tune(self, mock_from_id, mock_create):
+        ds_model = MagicMock()
+        ds_model.id = "test_id"
+        ds_model.compartment_id = "test_model_compartment_id"
+        ds_model.project_id = "test_project_id"
+        ds_model.display_name = "test_display_name"
+        ds_model.description = "test_description"
+        ds_model.model_version_set_id = "test_model_version_set_id"
+        ds_model.model_version_set_name = "test_model_version_set_name"
+        ds_model.freeform_tags = {
+            "license": "test_license",
+            "organization": "test_organization",
+            "task": "test_task",
+            "aqua_fine_tuned_model": "test_finetuned_model",
+        }
+        ds_model.time_created = "2024-01-19T17:57:39.158000+00:00"
+        ds_model.lifecycle_state = "ACTIVE"
+        custom_metadata_list = ModelCustomMetadata()
+        custom_metadata_list.add(
+            **{"key": "artifact_location", "value": "oci://bucket@namespace/prefix/"}
+        )
+        custom_metadata_list.add(
+            **{"key": "fine_tune_source", "value": "test_fine_tuned_source_id"}
+        )
+        custom_metadata_list.add(
+            **{"key": "fine_tune_source_name", "value": "test_fine_tuned_source_name"}
+        )
+        custom_metadata_list.add(
+            **{
+                "key": "deployment-container",
+                "value": "odsc-vllm-serving",
+            }
+        )
+        custom_metadata_list.add(
+            **{
+                "key": "evaluation-container",
+                "value": "odsc-llm-evaluate",
+            }
+        )
+        custom_metadata_list.add(
+            **{
+                "key": "finetune-container",
+                "value": "odsc-llm-fine-tuning",
+            }
+        )
+        ds_model.custom_metadata_list = custom_metadata_list
+        defined_metadata_list = ModelTaxonomyMetadata()
+        defined_metadata_list["Hyperparameters"].value = {
+            "training_data": "test_training_data",
+            "val_set_size": "test_val_set_size",
+        }
+        ds_model.defined_metadata_list = defined_metadata_list
+        ds_model.provenance_metadata = ModelProvenanceMetadata(
+            training_id="test_training_job_run_id"
+        )
+        ds_model.model_file_description = {
+            "version": "1.0",
+            "type": "modelOSSReferenceDescription",
+            "models": [
+                {
+                    "namespace": "test_namespace_one",
+                    "bucketName": "test_bucket_name_one",
+                    "prefix": "test_prefix_one",
+                    "objects": [
+                        {
+                            "name": "artifact/.gitattributes",
+                            "version": "123",
+                            "sizeInBytes": 1519,
+                        }
+                    ],
+                },
+                {
+                    "namespace": "test_namespace_two",
+                    "bucketName": "test_bucket_name_two",
+                    "prefix": "test_prefix_two",
+                    "objects": [
+                        {
+                            "name": "/README.md",
+                            "version": "b52c2608-009f-4774-8325-60ec226ae003",
+                            "sizeInBytes": 5189,
+                        }
+                    ],
+                },
+            ],
+        }
+
+        mock_from_id.return_value = ds_model
+
+        # missing 'OCI_AQUA' tag
+        with pytest.raises(
+            AquaValueError,
+            match="Model 'mock_model_id' is not eligible for conversion. Only legacy AQUA fine-tuned models without the 'fine_tune_model_version=v2' tag are supported.",
+        ):
+            self.app.convert_fine_tune(model_id="mock_model_id")
+
+        # add 'OCI_AQUA' tag
+        mock_from_id.return_value.freeform_tags["OCI_AQUA"] = "ACTIVE"
+
+        self.app.convert_fine_tune(model_id="mock_model_id")
+
+        mock_create.assert_called_with(model_by_reference=True)
+
+        assert mock_from_id.return_value.model_file_description["models"] == [
+            {
+                "namespace": "test_namespace_two",
+                "bucketName": "test_bucket_name_two",
+                "prefix": "test_prefix_two",
+                "objects": [
+                    {
+                        "name": "/README.md",
+                        "version": "b52c2608-009f-4774-8325-60ec226ae003",
+                        "sizeInBytes": 5189,
+                    }
+                ],
+            }
+        ]
+
     @pytest.mark.parametrize(
         ("artifact_location_set", "download_from_hf", "cleanup_model_cache"),
         [
