Solved bugs in finetuning vision yamls.

cli/foundation-models/system/finetune/image-classification/multiclass-classification/deploy.yaml

@@ -1,4 +1,11 @@
 $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
 name: demo
 instance_type: Standard_DS3_v2
-instance_count: 1
+instance_count: 1
+liveness_probe:
+  initial_delay: 180
+  period: 180
+  failure_threshold: 49
+  timeout: 299
+request_settings:
+  request_timeout_ms: 60000

cli/foundation-models/system/finetune/image-classification/multiclass-classification/hftransformers-fridgeobjects-multiclass-classification.sh

@@ -130,7 +130,7 @@ fi
 model_version=$(az ml model show --name $aml_registry_model_name --label $model_label --registry-name $registry_name --query version --output tsv)
 
 # 4. Prepare data
-python prepare_data.py
+python prepare_data.py --subscription $subscription_id --group $resource_group_name --workspace $workspace_name
 # training data
 train_data="./data/training-mltable-folder"
 # validation data

cli/foundation-models/system/finetune/image-classification/multiclass-classification/prepare_data.py

@@ -143,7 +143,7 @@ def read_image(image_path):
     parser = argparse.ArgumentParser(description="Prepare data for image classification")
 
     parser.add_argument("--subscription", type=str, help="Subscription ID")
-    parser.add_argument("--resource_group", type=str, help="Resource group name")
+    parser.add_argument("--group", type=str, help="Resource group name")
     parser.add_argument("--workspace", type=str, help="Workspace name")
     parser.add_argument("--data_path", type=str, default="./data", help="Dataset location")
 
@@ -165,8 +165,9 @@ def read_image(image_path):
 
     sample_image = os.path.join(args.data_path, "fridgeObjects", "milk_bottle", "99.jpg")
     huggingface_request_json = {
-        "inputs": {
-            "image": [base64.encodebytes(read_image(sample_image)).decode("utf-8")],
+        "input_data": {
+            "columns": ["image"],
+            "data": [base64.encodebytes(read_image(sample_image)).decode("utf-8")],
         }
     }
     huggingface_request_file_name = "huggingface_sample_request_data.json"

cli/foundation-models/system/finetune/image-classification/multilabel-classification/deploy.yaml

@@ -1,4 +1,11 @@
 $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
 name: demo
 instance_type: Standard_DS3_v2
-instance_count: 1
+instance_count: 1
+liveness_probe:
+  initial_delay: 180
+  period: 180
+  failure_threshold: 49
+  timeout: 299
+request_settings:
+  request_timeout_ms: 60000

cli/foundation-models/system/finetune/image-classification/multilabel-classification/hftransformers-fridgeobjects-multilabel-classification.sh

@@ -129,7 +129,7 @@ fi
 model_version=$(az ml model show --name $aml_registry_model_name --label $model_label --registry-name $registry_name --query version --output tsv)
 
 # 4. Prepare data
-python prepare_data.py
+python prepare_data.py --subscription $subscription_id --group $resource_group_name --workspace $workspace_name
 # training data
 train_data="./data/training-mltable-folder"
 # validation data

cli/foundation-models/system/finetune/image-classification/multilabel-classification/prepare_data.py

@@ -170,8 +170,9 @@ def read_image(image_path):
 
     sample_image = os.path.join(args.data_path, "multilabelFridgeObjects", "images", "56.jpg")
     huggingface_request_json = {
-        "inputs": {
-            "image": [base64.encodebytes(read_image(sample_image)).decode("utf-8")],
+        "input_data": {
+            "columns": ["image"],
+            "data": [base64.encodebytes(read_image(sample_image)).decode("utf-8")],
         }
     }
     huggingface_request_file_name = "huggingface_sample_request_data.json"

cli/foundation-models/system/finetune/image-instance-segmentation/deploy.yaml

@@ -1,4 +1,11 @@
 $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
 name: demo
 instance_type: Standard_DS3_v2
-instance_count: 1
+instance_count: 1
+liveness_probe:
+  initial_delay: 180
+  period: 180
+  failure_threshold: 49
+  timeout: 299
+request_settings:
+  request_timeout_ms: 60000

cli/foundation-models/system/finetune/image-instance-segmentation/mmdetection-fridgeobjects-instance-segmentation.sh

@@ -23,7 +23,7 @@ gpus_per_node=1
 # TODO: update the model name once it registered in preview registry
 # using the latest version of the model - not working yet
 mmdetection_model_name="mask_rcnn_swin-t-p4-w7_fpn_1x_coco"
-model_version=1
+model_label="latest"
 
 version=$(date +%s)
 finetuned_mmdetection_model_name="mask_rcnn_swin-t-p4-w7_fpn_1x_coco_fridge_is"
@@ -37,7 +37,6 @@ ds_finetune="./deepspeed_configs/zero1.json"
 mmdetection_sample_request_data="./mmdetection_sample_request_data.json"
 
 # finetuning job parameters
-# TODO: update with preview registry component name
 finetuning_pipeline_component="mmdetection_image_objectdetection_instancesegmentation_pipeline"
 
 # Training settings
@@ -123,14 +122,15 @@ fi
 
 # # 3. Check if the model exists in the registry
 # # need to confirm model show command works for registries outside the tenant (aka system registry)
-if ! az ml model show --name $model_name --version $model_version --registry-name $registry_name 
+if ! az ml model show --name $mmdetection_model_name --label $model_label --registry-name $registry_name 
 then
-    echo "Model $mmdetection_model_name:$model_version does not exist in registry $registry_name"
+    echo "Model $mmdetection_model_name:$model_label does not exist in registry $registry_name"
     exit 1
 fi
-
+# get the latest model version
+model_version=$(az ml model show --name $mmdetection_model_name --label $model_label --registry-name $registry_name --query version --output tsv)
 # 4. Prepare data
-python prepare_data.py
+python prepare_data.py --subscription $subscription_id --group $resource_group_name --workspace $workspace_name
 
 # training data
 train_data="./data/training-mltable-folder"
@@ -158,10 +158,10 @@ mmdetection_parent_job=$( az ml job create \
   --file ./mmdetection-fridgeobjects-instance-segmentation-pipeline.yml \
   $workspace_info \
   --set \
-  jobs.mmdetection_model_finetune_job.component=$finetuning_pipeline_component \
+  jobs.mmdetection_model_finetune_job.component="azureml://registries/$registry_name/components/$finetuning_pipeline_component/labels/latest" \
   inputs.compute_model_import=$compute_cluster_model_import \
   inputs.compute_finetune=$compute_cluster_finetune \
-  inputs.mlflow_model.path=$mmdetection_model_name \
+  inputs.mlflow_model.path="azureml://registries/$registry_name/models/$mmdetection_model_name/versions/$model_version" \
   inputs.training_data.path=$train_data \
   inputs.validation_data.path=$validation_data
   ) || {

cli/foundation-models/system/finetune/image-object-detection/deploy.yaml

@@ -1,4 +1,11 @@
 $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
 name: demo
 instance_type: Standard_DS3_v2
-instance_count: 1
+instance_count: 1
+liveness_probe:
+  initial_delay: 180
+  period: 180
+  failure_threshold: 49
+  timeout: 299
+request_settings:
+  request_timeout_ms: 60000

cli/foundation-models/system/finetune/image-object-detection/mmdetection-fridgeobjects-detection.sh

@@ -22,7 +22,7 @@ gpus_per_node=1
 # TODO: update the model name once it registered in preview registry
 # using the latest version of the model - not working yet
 mmdetection_model_name="yolof_r50_c5_8x8_1x_coco"
-model_version=1
+model_label="latest"
 
 version=$(date +%s)
 finetuned_mmdetection_model_name="yolof_r50_c5_8x8_1x_coco_fridge_od"
@@ -122,14 +122,15 @@ fi
 
 # # 3. Check if the model exists in the registry
 # # need to confirm model show command works for registries outside the tenant (aka system registry)
-if ! az ml model show --name $model_name --version $model_version --registry-name $registry_name 
+if ! az ml model show --name $mmdetection_model_name --label $model_label --registry-name $registry_name 
 then
-    echo "Model $mmdetection_model_name:$model_version does not exist in registry $registry_name"
+    echo "Model $mmdetection_model_name:$model_label does not exist in registry $registry_name"
     exit 1
 fi
-
+# get the latest model version
+model_version=$(az ml model show --name $mmdetection_model_name --label $model_label --registry-name $registry_name --query version --output tsv)
 # 4. Prepare data
-python prepare_data.py
+python prepare_data.py --subscription $subscription_id --group $resource_group_name --workspace $workspace_name
 
 # training data
 train_data="./data/training-mltable-folder"
@@ -157,10 +158,10 @@ mmdetection_parent_job=$( az ml job create \
   --file ./mmdetection-fridgeobjects-detection-pipeline.yml \
   $workspace_info \
   --set \
-  jobs.mmdetection_model_finetune_job.component=$finetuning_pipeline_component \
+  jobs.mmdetection_model_finetune_job.component="azureml://registries/$registry_name/components/$finetuning_pipeline_component/labels/latest" \
   inputs.compute_model_import=$compute_cluster_model_import \
   inputs.compute_finetune=$compute_cluster_finetune \
-  inputs.mlflow_model.path=$mmdetection_model_name \
+  inputs.mlflow_model.path="azureml://registries/$registry_name/models/$mmdetection_model_name/versions/$model_version" \
   inputs.training_data.path=$train_data \
   inputs.validation_data.path=$validation_data
   ) || {