Merge branch 'main' into add-appinsights-logging

cli/foundation-models/system/inference/text-to-image/batch-deploy.yml

@@ -1,5 +1,5 @@
 $schema: https://azuremlschemas.azureedge.net/latest/batchDeployment.schema.json
-description: "Batch endpoint for for text-to-image task"
+description: "Batch endpoint for for (image-)text-to-image task"
 type: model
 resources:
     instance_count: 1

cli/foundation-models/system/inference/text-to-image/image-text-to-image-batch-endpoint.sh

@@ -0,0 +1,113 @@
+set -x
+# the commands in this file map to steps in this notebook: https://aka.ms/azureml-infer-batch-sdk-text-classification
+
+# script inputs
+registry_name="azureml"
+
+subscription_id="<SUBSCRIPTION_ID>"
+resource_group_name="<RESOURCE_GROUP>"
+workspace_name="<WORKSPACE_NAME>"
+
+# This is the model from system registry that needs to be deployed
+model_name="stabilityai-stable-diffusion-xl-refiner-1-0"
+model_label="latest"
+
+version=$(date +%s)
+endpoint_name="image-text-to-image-$version"
+deployment_name="image-text-to-image-batch-deploy"
+
+deployment_compute="gpu-cluster"
+compute_sku="Standard_NC6s_v3"
+
+# 1. Setup pre-requisites
+if [ "$subscription_id" = "<SUBSCRIPTION_ID>" ] || \
+   ["$resource_group_name" = "<RESOURCE_GROUP>" ] || \
+   [ "$workspace_name" = "<WORKSPACE_NAME>" ]; then 
+    echo "Please update the script with the subscription_id, resource_group_name and workspace_name"
+    exit 1
+fi
+
+az account set -s $subscription_id
+workspace_info="--resource-group $resource_group_name --workspace-name $workspace_name"
+
+# 2. 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 --label $model_label --registry-name $registry_name 
+then
+    echo "Model $model_name:$model_label does not exist in registry $registry_name"
+    exit 1
+fi
+
+model_version=$(az ml model show --name $model_name --label $model_label --registry-name $registry_name --query version --output tsv)
+
+# 3. Check if compute $deployment_compute exists, else create it
+if az ml compute show --name $deployment_compute $workspace_info
+then
+    echo "Compute cluster $deployment_compute already exists"
+else
+    echo "Creating compute cluster $deployment_compute"
+    az ml compute create --name $deployment_compute --type amlcompute --min-instances 0 --max-instances 2 --size $compute_sku $workspace_info || {
+        echo "Failed to create compute cluster $deployment_compute"
+        exit 1
+    }
+fi
+
+# 4. Deploy the model to an endpoint
+# create batch endpoint 
+az ml batch-endpoint create --name $endpoint_name $workspace_info  || {
+    echo "endpoint create failed"; exit 1;
+}
+
+# deploy model from registry to endpoint in workspace
+az ml batch-deployment create --file batch-deploy.yml $workspace_info --set \
+  endpoint_name=$endpoint_name \
+  name=$deployment_name \
+  compute=$deployment_compute \
+  model=azureml://registries/$registry_name/models/$model_name/versions/$model_version || {
+    echo "deployment create failed"; exit 1;
+}
+
+# 4. Submit a sample request to endpoint
+data_path="./inpainting_data/batch_data"
+python utils/prepare_data_image_text_to_image.py --payload-path $data_path --mode "batch"
+# Path where the processes csvs are dumped. This is the input to the endpoint
+processed_data_path="./inpainting_data/batch_data/processed_batch_data"
+
+# Check if scoring folder exists
+if [ -d $processed_data_path ]; then
+    echo "Invoking endpoint $endpoint_name with following input:\n\n"
+    ls $processed_data_path
+    echo "\n\n"
+else
+    echo "Scoring folder $processed_data_path does not exist"
+    exit 1
+fi
+
+# 5. Invoke a job on the batch endpoint
+job_name=$(az ml batch-endpoint invoke --name $endpoint_name \
+ --deployment-name $deployment_name \
+ --input $processed_data_path \
+ --input-type uri_folder --query name --output tsv $workspace_info) || {
+    echo "endpoint invoke failed"; exit 1;
+}
+
+# 6. Stream the job logs
+az ml job stream --name $job_name $workspace_info || {
+    echo "job stream-logs failed. If the job failed with Assertion Error stating actual size of csv exceeds 100 MB, then try splitting input csv file into multiple csv files."; exit 1;
+}
+
+# 7. Download the job output
+az ml job download --name $job_name --download-path "generated_images" $workspace_info || {
+    echo "job output download failed"; exit 1;
+}
+
+# 8. Delete the endpoint
+az ml batch-endpoint delete --name $endpoint_name $workspace_info --yes || {
+    echo "endpoint delete failed"; exit 1;
+}
+
+# 9. Delete the compute cluster (Uncomment the below lines to delete the created cluster)
+# az ml compute delete --name $deployment_compute $workspace_info --yes || {
+#     echo "compute delete failed"; exit 1;
+# }
+

cli/foundation-models/system/inference/text-to-image/image-text-to-image-online-endpoint.sh

@@ -0,0 +1,83 @@
+set -x
+# the commands in this file map to steps in this notebook: https://aka.ms/azureml-infer-batch-sdk-text-classification
+
+# script inputs
+registry_name="azureml"
+
+subscription_id="<SUBSCRIPTION_ID>"
+resource_group_name="<RESOURCE_GROUP>"
+workspace_name="<WORKSPACE_NAME>"
+
+# This is the model from system registry that needs to be deployed
+model_name="stabilityai-stable-diffusion-xl-refiner-1-0"
+model_label="latest"
+response_file="generated_image.json"
+
+version=$(date +%s)
+endpoint_name="image-text-to-image-$version"
+deployment_name="image-text-to-image-deploy"
+
+deployment_sku="Standard_NC6s_v3"
+
+# sample_request_data
+sample_request_data="inpainting_data/sample_request_data.json"
+
+# 1. Setup pre-requisites
+if [ "$subscription_id" = "<SUBSCRIPTION_ID>" ] || \
+   ["$resource_group_name" = "<RESOURCE_GROUP>" ] || \
+   [ "$workspace_name" = "<WORKSPACE_NAME>" ]; then 
+    echo "Please update the script with the subscription_id, resource_group_name and workspace_name"
+    exit 1
+fi
+
+az account set -s $subscription_id
+workspace_info="--resource-group $resource_group_name --workspace-name $workspace_name"
+
+# 2. Check if the model exists in the registry
+
+if ! az ml model show --name $model_name --label $model_label --registry-name $registry_name
+then
+    echo "Model $model_name:$model_label does not exist in registry $registry_name"
+    exit 1
+fi
+
+# Get the latest model version
+model_version=$(az ml model show --name $model_name --label $model_label --registry-name $registry_name --query version --output tsv)
+
+# 3. Deploy the model to an endpoint
+# Create online endpoint 
+az ml online-endpoint create --name $endpoint_name $workspace_info  || {
+    echo "endpoint create failed"; exit 1;
+}
+
+# Deploy model from registry to endpoint in workspace
+az ml online-deployment create --file deploy-online.yaml $workspace_info --all-traffic --set \
+  endpoint_name=$endpoint_name model=azureml://registries/$registry_name/models/$model_name/versions/$model_version \
+  name=$deployment_name \
+  instance_type=$deployment_sku || {
+    echo "deployment create failed"; exit 1;
+}
+
+# 4. Submit a sample request to endpoint
+python utils/prepare_data_image_text_to_image.py --payload-path $sample_request_data --mode "online"
+
+# Check if scoring data file exists
+if [ -f $sample_request_data ]; then
+    echo "Invoking endpoint $endpoint_name with $sample_request_data:\n\n"
+    echo "\n\n"
+else
+    echo "Request file $sample_request_data does not exist"
+    exit 1
+fi
+
+az ml online-endpoint invoke --name $endpoint_name --request-file $sample_request_data $workspace_info  -o json > $response_file || {
+    echo "endpoint invoke failed"; exit 1;
+}
+
+# 5. Convert bas64 form of image to jpeg
+python ./utils/base64_to_jpeg.py --response_file $response_file
+
+# 6. Delete the endpoint and sample_request_data.json
+az ml online-endpoint delete --name $endpoint_name $workspace_info --yes || {
+    echo "endpoint delete failed"; exit 1;
+}

cli/foundation-models/system/inference/text-to-image/utils/prepare_data_image_text_to_image.py

@@ -0,0 +1,151 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+# Prepare request payload for the image-text to image task
+
+import argparse
+import glob
+import json
+import io
+import base64
+import os
+import pandas as pd
+from PIL import Image
+from pathlib import Path
+
+
+def read_image(image_path: str) -> bytes:
+    """Reads an image from a file path into a byte array.
+
+    :param image_path: Path to image file.
+    :type image_path: str
+    :return: Byte array of image.
+    :rtype: bytes
+    """
+    with open(image_path, "rb") as f:
+        return f.read()
+
+
+def prepare_batch_payload(payload_path: str) -> None:
+    """Prepare payload for online deployment.
+
+    :param payload_path: Path to payload csv file.
+    :type payload_path: str
+    :return: None
+    """
+
+    base_image1 = "inpainting_data/images/dog_on_bench.png"
+    base_image2 = "inpainting_data/images/teapot.png"
+
+    os.makedirs(payload_path, exist_ok=True)
+
+    input_data = {
+        "columns": ["image", "prompt"],
+        "data": [
+            {
+                "image": base64.encodebytes(read_image(base_image1)).decode("utf-8"),
+                "prompt": "A yellow cat, high resolution, sitting on a park bench",
+            },
+            {
+                "image": base64.encodebytes(read_image(base_image2)).decode("utf-8"),
+                "prompt": "A small flower featuring a blend of pink and purple colors.",
+            },
+        ],
+    }
+    pd.DataFrame(**input_data).to_csv(
+        os.path.join(payload_path, "input1.csv"), index=False
+    )
+
+    input_data = {
+        "columns": ["image", "prompt"],
+        "data": [
+            {
+                "image": base64.encodebytes(read_image(base_image1)).decode("utf-8"),
+                "prompt": "Pikachu, cinematic, digital art, sitting on bench",
+            },
+            {
+                "image": base64.encodebytes(read_image(base_image2)).decode("utf-8"),
+                "prompt": "A woman with red hair in the style of Tamara de Lempicka.",
+            },
+        ],
+    }
+    pd.DataFrame(**input_data).to_csv(
+        os.path.join(payload_path, "input2.csv"), index=False
+    )
+
+    # Use glob to get a list of CSV files in the folder
+    csv_files = glob.glob(os.path.join(payload_path, "*.csv"))
+
+    # Read all CSV files into a single DataFrame using pd.concat
+    batch_df = pd.concat((pd.read_csv(file) for file in csv_files), ignore_index=True)
+
+    # Specify the folder where your CSV files should be saved
+    processed_dataset_parent_dir = os.path.join(payload_path, "processed_batch_data")
+    os.makedirs(processed_dataset_parent_dir, exist_ok=True)
+    batch_input_file = "batch_input.csv"
+
+    # Divide this into files of <x> rows each
+    batch_size_per_predict = 2
+    for i in range(0, len(batch_df), batch_size_per_predict):
+        j = i + batch_size_per_predict
+        batch_df[i:j].to_csv(
+            os.path.join(processed_dataset_parent_dir, str(i) + batch_input_file)
+        )
+
+    # Check out the first and last file name created
+    input_paths = sorted(
+        Path(processed_dataset_parent_dir).iterdir(), key=os.path.getmtime
+    )
+    input_files = [os.path.basename(path) for path in input_paths]
+    print(f"{input_files[0]} to {str(i)}{batch_input_file}.")
+
+
+def prepare_online_payload(payload_path: str) -> None:
+    """Prepare payload for online deployment.
+
+    :param payload_path: Path to payload json file.
+    :type payload_path: str
+    :return: None
+    """
+    base_directory = os.path.dirname(os.path.dirname(__file__))
+
+    base_image = os.path.join(
+        base_directory, "inpainting_data", "images", "dog_on_bench.png"
+    )
+
+    request_json = {
+        "input_data": {
+            "columns": ["image", "prompt"],
+            "index": [0],
+            "data": [
+                {
+                    "image": base64.encodebytes(read_image(base_image)).decode("utf-8"),
+                    "prompt": "A yellow cat, high resolution, sitting on a park bench",
+                }
+            ],
+        }
+    }
+    print(base_directory)
+    print(payload_path)
+    payload_path = os.path.join(base_directory, payload_path)
+    with open(payload_path, "w") as request_file:
+        json.dump(request_json, request_file)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Prepare sample data for image-text to image"
+    )
+    parser.add_argument("--payload-path", type=str, help="payload file/ folder path")
+    parser.add_argument(
+        "--mode",
+        type=str,
+        default="online",
+        help="Generate payload for online or batch deployment.",
+    )
+    args, unknown = parser.parse_known_args()
+
+    if args.mode == "online":
+        prepare_online_payload(args.payload_path)
+    else:
+        prepare_batch_payload(args.payload_path)

sdk/python/foundation-models/system/inference/text-to-image/aacs-scoring-files/docker_env/conda_dependencies.yaml

@@ -7,7 +7,7 @@ dependencies:
   - mlflow==2.3.2
   - torch==1.13.0
   - transformers==4.29.1
-  - diffusers==0.20.0
+  - diffusers==0.23.0
   - accelerate==0.22.0
   - azureml-core==1.52.0
   - azureml-mlflow==1.52.0