Bugfix: use correct default model and set intel gpu to true (#1035)

Author: maxxgx

interactive_ai/libs/iai_core_py/iai_core/services/model_service.py

@@ -178,16 +178,16 @@ def _get_active_model_state(
     def get_or_create_model_storage(
         project_identifier: ProjectIdentifier,
         task_node: TaskNode,
-        model_template_id: str,
+        model_manifest_id: str,
     ) -> ModelStorage:
         """
-        Returns the model storage for a particular task_node and model_template. If no
-        model storage for that task node and model template exists, this method will
+        Returns the model storage for a particular task_node and model_manifest. If no
+        model storage for that task node and model manifest exists, this method will
         create one.
 
         :param project_identifier: Identifier of the project containing the task node
         :param task_node: Task node associated with the model storage to get
-        :param model_template_id: Identifier of the model template associated with
+        :param model_manifest_id: Identifier of the model manifest associated with
             the model storage to get
         :return: ModelStorage
         """
@@ -196,10 +196,10 @@ def get_or_create_model_storage(
         model_storages = model_storage_repo.get_by_task_node_id(task_node_id=task_node.id_)
 
         for task_model_storage in model_storages:
-            if task_model_storage.model_template.model_template_id == model_template_id:
+            if task_model_storage.model_manifest_id == model_manifest_id:
                 model_storage = task_model_storage
         if model_storage is None:
-            model_template = ModelTemplateList().get_by_id(model_template_id)
+            model_template = ModelTemplateList().get_by_id(model_manifest_id)
             if (
                 isinstance(model_template, NullModelTemplate)
                 or model_template.task_type != task_node.task_properties.task_type
@@ -210,7 +210,7 @@ def get_or_create_model_storage(
                     if model_template.task_type == task_node.task_properties.task_type
                 ]
                 raise ValueError(
-                    f"Algorithm with name '{model_template_id}' was not found for task "
+                    f"Algorithm with name '{model_manifest_id}' was not found for task "
                     f"{task_node.title} of type {task_node.task_properties.task_type}. "
                     f"Algorithms that are available to this task are: {available_algos}."
                 )

interactive_ai/libs/iai_core_py/tests/services/test_model_service.py

@@ -109,29 +109,29 @@ def test_get_or_create_model_storage(
         model_storage = ModelService.get_or_create_model_storage(
             project_identifier=fxt_empty_project_persisted.identifier,
             task_node=fxt_detection_task,
-            model_template_id=fxt_model_template_detection.model_template_id,
+            model_manifest_id=fxt_model_template_detection.model_template_id,
         )
         request.addfinalizer(lambda: model_storage_repo.delete_by_id(model_storage.id_))
         # Test case 2: Model storage retrieval
         model_storage_exists = ModelService.get_or_create_model_storage(
             project_identifier=fxt_empty_project_persisted.identifier,
             task_node=fxt_detection_task,
-            model_template_id=fxt_model_template_detection.model_template_id,
+            model_manifest_id=fxt_model_template_detection.model_template_id,
         )
         # Test case 3: Raise error for invalid model template id
         invalid_template_id = "non existing model template id"
         with pytest.raises(ValueError) as exc_non_existing:
             ModelService.get_or_create_model_storage(
                 project_identifier=fxt_empty_project_persisted.identifier,
                 task_node=fxt_detection_task,
-                model_template_id=invalid_template_id,
+                model_manifest_id=invalid_template_id,
             )
         # Test case 4: Raise error for model template with non-matching task type
         with pytest.raises(ValueError) as exc_non_matching:
             ModelService.get_or_create_model_storage(
                 project_identifier=fxt_empty_project_persisted.identifier,
                 task_node=fxt_detection_task,
-                model_template_id=fxt_model_template_classification.model_template_id,
+                model_manifest_id=fxt_model_template_classification.model_template_id,
             )
 
         # Assert

interactive_ai/services/director/app/communication/controllers/training_controller.py

@@ -180,7 +180,7 @@ def _submit_train_job(project: Project, task_training_config: TrainingConfig, au
         model_storage = ModelService.get_or_create_model_storage(
             project_identifier=project.identifier,
             task_node=task_node,
-            model_template_id=task_training_config.model_template_id,
+            model_manifest_id=task_training_config.model_template_id,
         )
 
         logger.info("Submitting train job for task with ID `%s`", task_node.id_)

interactive_ai/services/director/app/configuration/configuration_manager.py

@@ -275,7 +275,7 @@ def get_configuration_for_algorithm(
             model_storage = ModelService.get_or_create_model_storage(
                 project_identifier=project.identifier,
                 task_node=trainable_task,
-                model_template_id=algorithm_name,
+                model_manifest_id=algorithm_name,
             )
         except ValueError as error:
             raise AlgorithmNotFoundException(message=str(error))

interactive_ai/services/director/app/usecases/auto_train/kafka_handler.py

@@ -4,14 +4,15 @@
 
 import logging
 
+from geti_supported_models.default_models import DefaultModels
+
 from coordination.dataset_manager.dynamic_required_num_annotations import DynamicRequiredAnnotations
 from environment import get_gpu_provider
 
 from .auto_train_use_case import AutoTrainUseCase
 from geti_kafka_tools import BaseKafkaHandler, KafkaRawMessage, TopicSubscription
 from geti_telemetry_tools import unified_tracing
 from geti_types import ID, ProjectIdentifier, Singleton
-from iai_core.algorithms import ModelTemplateList
 from iai_core.repos import ProjectRepo
 from iai_core.services import ModelService
 from iai_core.session.session_propagation import setup_session_kafka
@@ -61,23 +62,19 @@ def on_project_created(self, raw_message: KafkaRawMessage) -> None:
         for task_node in project.get_trainable_task_nodes():
             task_type = task_node.task_properties.task_type
             try:
-                default_model_template_id = next(
-                    model_template.model_template_id
-                    for model_template in ModelTemplateList().get_all()
-                    if model_template.task_type == task_type and model_template.is_default_for_task
-                )
-            except StopIteration:
-                logger.error("Could not resolve default model template for task type %s", task_type)
+                default_model_manifest_id = DefaultModels.get_default_model(task_type.name)
+            except ValueError:
+                logger.error("Could not resolve default model manifest for task type %s", task_type)
                 return
             logger.info(
                 "[WORKAROUND] Switching active model storage to '%s' for task '%s'",
-                default_model_template_id,
+                default_model_manifest_id,
                 task_node,
             )
             model_storage = ModelService.get_or_create_model_storage(
                 project_identifier=project_identifier,
                 task_node=task_node,
-                model_template_id=default_model_template_id,
+                model_manifest_id=default_model_manifest_id,
             )
             ModelService.activate_model_storage(model_storage)
 

interactive_ai/services/director/tests/unit/communication/rest_controllers/test_training_controller.py

@@ -306,7 +306,7 @@ def test_submit_train_jobs_new_model_storage(
         # Assert
         patched_get_model_storage.assert_called_once_with(
             task_node=fxt_detection_task,
-            model_template_id=fxt_model_template_detection.model_template_id,
+            model_manifest_id=fxt_model_template_detection.model_template_id,
             project_identifier=fxt_project.identifier,
         )
         mock_training_job_submit.assert_called_once_with(

interactive_ai/services/director/tests/unit/configuration/test_configuration_manager.py

@@ -107,7 +107,7 @@ def test_get_configuration_for_algorithm_uninitialized_model_storage(
         mock_save_hyper_parameters.assert_called_once_with(hyper_parameters)
         mock_get_model_storage.assert_called_once_with(
             task_node=detection_task,
-            model_template_id=algorithm_name,
+            model_manifest_id=algorithm_name,
             project_identifier=fxt_project_with_detection_task.identifier,
         )
 
@@ -125,7 +125,7 @@ def test_get_configuration_for_algorithm_initialized_model_storage(
     ):
         # Arrange
         detection_task = fxt_project_with_detection_task.tasks[-1]
-        algorithm_name = "test_model_template_id"
+        algorithm_name = fxt_model_template_detection.model_manifest_id
 
         # Set a non-default value for one of the parameters to make sure we're not
         # just returning the defaults, upon comparison later on
@@ -138,7 +138,6 @@ def test_get_configuration_for_algorithm_initialized_model_storage(
                 id_to_str=True,
             )
         )
-        fxt_model_template_detection.model_template_id = algorithm_name
         fxt_model_storage_detection._model_template = fxt_model_template_detection
 
         # Act

interactive_ai/supported_models/geti_supported_models/manifests/detection/dfine_x.yaml

@@ -10,9 +10,6 @@ stats:
     training_time: 1
     inference_speed: 1
 
-supported_gpus:
-  intel: false
-
 hyperparameters:
   training:
     learning_rate: 0.00025

interactive_ai/supported_models/geti_supported_models/manifests/instance_segmentation/maskrcnn_efficientnetb2b.yaml

@@ -10,9 +10,6 @@ stats:
     training_time: 3
     inference_speed: 3
 
-supported_gpus:
-  intel: false
-
 hyperparameters:
   training:
     learning_rate: 0.007

interactive_ai/supported_models/geti_supported_models/manifests/instance_segmentation/maskrcnn_r50_v1.yaml

@@ -11,9 +11,6 @@ stats:
     inference_speed: 2
 support_status: deprecated
 
-supported_gpus:
-  intel: false
-
 hyperparameters:
   training:
     learning_rate: 0.007