Function Compute:Best practices for AI

Scenarios and benefits

Various issues often occur in the traditional GPU infrastructure for AI applications. These issues include long construction cycles, high O&M complexity, low cluster utilization, and high costs. GPU-accelerated instances of Function Compute hand over the issues from users to the cloud vendors. This way, you can focus on your business and achieve your business goals without the need to worry about the underlying GPU infrastructure.

This section describes the benefits of GPU-accelerated instances compared with the instances with no GPU acceleration in Function Compute:

• Cost prioritized AI application scenarios

  • Reserve GPU-accelerated instances based on their business requirements. This way, GPU-accelerated instances of Function Compute offer much higher cost efficiency than self-built GPU clusters.

  • Use GPU resources in 1/2 or exclusive mode through the GPU virtualization technology. This way, GPU-accelerated instances can be configured in a fine-grained manner.

• Efficiency prioritized AI application scenarios

  • Focus on code development and the business objectives without the need to perform O&M on GPU clusters, such as driver and CUDA version management, machine operation management, and faulty GPU management.

For more information about GPU-accelerated instances, see Instance types and instance modes.

Tutorial for neural style transfer

Neural style transfer (NST) is a technology that is used to blend two images. During this process, the content is extracted from one image and style is extracted from the other image to create a new image. In the following example, the TensorFlow Hub preset model is used to complete the stylistic synthesis of an image.

Synthesis effect

Before you start

• General preparations

  • For optimal user experience, join the DingTalk group (ID 11721331) and provide the following information:

    • The organization name, such as the name of your company.

    • The ID of your Alibaba Cloud account.

    • The region where you want to use GPU-accelerated instances, such as China (Shenzhen).

    • The contact information, such as your mobile number, email address, or DingTalk account.

  • Upload the audio and video resources that you want to process to an Object Storage Service (OSS) bucket in the region where the GPU-accelerated instances are located. Make sure that you have the read and write permissions on the objects in the bucket. For more information about how to upload audio and video resources, see Upload objects. For more information about permissions, see Modify the ACL of a bucket.

• GPU application deployment by using Serverless Devs

  • Perform the following operations in the region where the GPU-accelerated instances reside:

    • Create a Container Registry Enterprise Edition instance or Personal Edition instance. We recommend that you create an Enterprise Edition instance. For more information, see Step 1: Create a Container Registry Enterprise Edition instance.

    • Create a namespace and an image repository. For more information, see Step 2: Create a namespace and Step 3: Create an image repository.

  • Install Serverless Devs and Docker
  • Configure Serverless Devs

Use Serverless Devs to deploy GPU applications

1. Create a project.

   s init devsapp/start-fc-custom-container-event-python3.9 -d fc-gpu-prj

   The following sample code shows the directory of the created project:

   fc-gpu-prj
   ├── code
   │   ├── app.py        # Function code.
   │   └── Dockerfile    # Dockerfile: The image Dockerfile that contains the code.
   ├── README.md
   └── s.yaml            # Project configurations, which specify how the image is deployed in Function Compute

2. Go to the project directory.

   cd fc-gpu-prj

3. Modify the configurations of the directory file based on your business requirements.

   • Edit the s.yaml file.

     For more information about the parameters in the YAML file, see YAML specifications.

     edition: 1.0.0
     name: container-demo
     access: default
     vars:
       region: cn-shenzhen
     services:
       customContainer-demo:
         component: devsapp/fc
         props:
           region: ${vars.region}
           service:
             name: tgpu_tf_service
             internetAccess: true
           function:
             name: tgpu_tf_func
             description: test gpu for tensorflow
             handler: not-used
             timeout: 600
             caPort: 9000
             instanceType: fc.gpu.tesla.1
             gpuMemorySize: 8192
             cpu: 4
             memorySize: 16384
             diskSize: 512
             runtime: custom-container
             customContainerConfig:
               #1. Make sure that the namespace:demo namespace and the repo:gpu-tf-style-transfer_s repository are created in advance in Alibaba Cloud Container Registry. 
               #2. Change the tag from v0.1 to v0.2 when you update the function and run s build && s deploy again. 
               image: registry.cn-shenzhen.aliyuncs.com/demo/gpu-tf-style-transfer_s:v0.1
             codeUri: ./code
           triggers:
             - name: httpTrigger
               type: http
               config:
                 authType: anonymous
                 methods:
                   - GET

   • Edit the app.py file.

     Example:

     # -*- coding: utf-8 -*-
     # python2 and python3
     from __future__ import print_function
     from http.server import HTTPServer, BaseHTTPRequestHandler
     from tensorflow import keras
     from tensorflow.keras import layers
     from tensorflow.keras.models import Sequential
     import json
     import sys
     import logging
     import os
     import matplotlib.pyplot as plt
     import matplotlib as mpl
     import numpy as np
     import os
     import PIL
     import tensorflow as tf
     import pathlib
     import urllib.request
     import random
     
     class Resquest(BaseHTTPRequestHandler):
         def upload(self, url, path):
             print("enter upload:", url)
             headers = {
                 'Content-Type': 'application/octet-stream',
                 'Content-Length': os.stat(path).st_size,
             }
             req = urllib.request.Request(url, open(path, 'rb'), headers=headers, method='PUT')
             urllib.request.urlopen(req)
     
         def tensor_to_image(self, tensor):
             tensor = tensor*255
             tensor = np.array(tensor, dtype=np.uint8)
             if np.ndim(tensor)>3:
                 assert tensor.shape[0] == 1
                 tensor = tensor[0]
             return PIL.Image.fromarray(tensor)
     
         def load_img(self, path_to_img):
             max_dim = 512
             img = tf.io.read_file(path_to_img)
             img = tf.image.decode_image(img, channels=3)
             img = tf.image.convert_image_dtype(img, tf.float32)
     
             shape = tf.cast(tf.shape(img)[:-1], tf.float32)
             long_dim = max(shape)
             scale = max_dim / long_dim
     
             new_shape = tf.cast(shape * scale, tf.int32)
     
             img = tf.image.resize(img, new_shape)
             img = img[tf.newaxis, :]
             return img
     
         def do_style_transfer(self):
             mpl.rcParams['figure.figsize'] = (12,12)
             mpl.rcParams['axes.grid'] = False
     
             # Use the path of the Object Storage Service (OSS) object in your Alibaba Cloud account. You must have the read and write permissions on the object. 
             # Read the content and style images from your OSS buckets. 
             content_path = tf.keras.utils.get_file(str(random.randint(0,100000000)) + ".jpg", 'https://your_public_oss/c1.png')
             style_path = tf.keras.utils.get_file(str(random.randint(0,100000000)) + ".jpg",'https://your_public_oss/c2.png')
     
             content_image = self.load_img(content_path)
             style_image = self.load_img(style_path)
             print("load image ok")
     
             import tensorflow_hub as hub
             hub_model = hub.load('https://hub.tensorflow.google.cn/google/magenta/arbitrary-image-stylization-v1-256/2')
             # You can package the hub model into an image for loading to accelerate the processing.
             #hub_model = hub.load('/usr/src/app/style_transfer_model')
             stylized_image = hub_model(tf.constant(content_image), tf.constant(style_image))[0]
             print("load model ok")
     
             path = "/tmp/" + str(random.randint(0,100000000)) + ".png"
             self.tensor_to_image(stylized_image).save(path)
             print("generate stylized image ok")
     
             # Use the path of the Object Storage Service (OSS) object in your Alibaba Cloud account. You must have the read and write permissions on the object. 
             # Save the synthesized images to the OSS bucket. 
             self.upload("https://your_public_oss/stylized-image.png" ,path)
             return "transfer ok"
     
         def style_transfer(self):
             msg = self.do_style_transfer()
             data = {"result": msg}
             self.send_response(200)
             self.send_header("Content-type", "application/json")
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def pong(self):
             data = {"function":"tf_style_transfer"}
             self.send_response(200)
             self.send_header('Content-type', 'application/json')
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def dispatch(self):
             mode = self.headers.get('RUN-MODE')
     
             if mode == "ping":
                 self.pong()
             elif mode == "normal":
                 self.style_transfer()
             else:
                 self.pong()
     
         def do_GET(self):
             self.dispatch()
     
         def do_POST(self):
             self.dispatch()
     
     if __name__ == "__main__":
         host = ("0.0.0.0", 9000)
         server = HTTPServer(host, Resquest)
         print("Starting server, listen at: %s:%s" % host)
         server.serve_forever()

   • Edit the Dockerfile file.

     Example:

     FROM registry.cn-shanghai.aliyuncs.com/serverless_devs/tensorflow:2.7.0-gpu
     WORKDIR /usr/src/app
     RUN apt-get update
     RUN apt-get install -y python3
     RUN apt-get install -y python3-pip
     RUN pip3 install matplotlib
     RUN pip install tensorflow_hub
     COPY . .
     CMD [ "python3", "-u", "/usr/src/app/app.py" ]
     EXPOSE 9000

4. Build an image.

   s build --dockerfile ./code/Dockerfile

5. Deploy the code to Function Compute.

   s deploy

   Note

   If you run the preceding command repeatedly and service name and function name remain unchanged, run the use local command to use local configurations.

6. Configure provisioned instances.

   s provision put --target 1 --qualifier LATEST

7. Check whether the provisioned instances are ready.

   s provision get --qualifier LATEST

   If the value of current is 1, the provisioned mode of GPU-accelerated instances is ready. Example:

   [2022-06-21 11:53:19] [INFO] [FC] - Getting provision: tgpu_tf_service.LATEST/tgpu_tf_func
   helloworld:
     serviceName:            tgpu_tf_service
     functionName:           tgpu_tf_func
     qualifier:              LATEST
     resource:               188077086902****#tgpu_tf_service#LATEST#tgpu_tf_func
     target:                 1
     current:                1
     scheduledActions:
       (empty array)
     targetTrackingPolicies:
       (empty array)
     currentError:
     alwaysAllocateCPU:      true

8. Invoke the function.

   • View online function versions

     s invoke
     FC Invoke Result:
     {"function": "tf_style_transfer"}

   • Perform neural style transfer

     s invoke -e '{"method":"GET","headers":{"RUN-MODE":"normal"}}'
     generate stylized image ok
     enter upload: https://your_public_oss/stylized-image.png  # You can download this file to view the synthesis effect.
     FC Invoke Result:
     {"result": "transfer ok"}

9. Release the GPU-accelerated instances.

   s provision put --target 0 --qualifier LATEST

Use the Function Compute console to deploy a GPU-accelerated application

1. Deploy an image.

   1. Create a Container Registry Enterprise Edition instance or Container Registry Personal Edition instance.

      We recommend that you create an Enterprise Edition instance. For more information, see Create a Container Registry Enterprise Edition instance.

   2. Create a namespace and an image repository.

      For more information, see the Step 2: Create a namespace and Step 3: Create an image repository sections of the "Use Container Registry Enterprise Edition instances to build images" topic.

   3. Perform operations on Docker as prompted in the Container Registry console. Then, push the preceding sample app.py and Dockerfile to the instance image repository. For more information about the files, see app.py and Dockerfile in the /code directory when you deploy a GPU application by using Serverless Devs.

      

2. Create a service. For more information, see the "Create a service“ section of the Manage services section.

3. Create a function. For more information, see Create a custom container function.

   Note

   Select GPU Instance for Instance Type and Process HTTP Requests for Request Handler Type.

4. Change the execution timeout period of the function.

   1. Find the function that you want to manage and click Configure in the Actions column.

   2. In the Environment Variables section, change the value of Execution Timeout Period and click Save.

      

   Note

   Transcoding duration by using CPU exceeds the default value of 60 seconds. Therefore, we recommend that you set the value of Execution Timeout Period to a larger value.

5. Configure a provisioned GPU-accelerated instance.

   1. On the function details page, click the Auto Scaling tab and click Create Rule.

   2. On the page that appears, configure the following parameters to provision GPU-accelerated instances and click Create.

      For more information about how to configure provisioned instances, see Configure provisioned instances and auto scaling rules.

   After the configuration is complete, you can check whether the provisioned GPU-accelerated instances are ready in the rule list. Specifically, check whether the value of Current Reserved Instances is the specified number of provisioned instances.

6. Use cURL to test the function.

   1. On the function details page, click the Triggers tab to view trigger configurations and obtain the trigger endpoint.

   2. Run the following command in the CLI to invoke the GPU-accelerated function:

      • View online function versions

        curl -v "https://tgpu-ff-console-tgpu-ff-console-ajezot****.cn-shenzhen.fcapp.run"
        {"function": "trans_gpu"}

      • Perform neural style transfer

        curl "https://tgpu-fu-console-tgpu-se-console-zpjido****.cn-shenzhen.fcapp.run" -H 'RUN-MODE: normal'
        {"result": "transfer ok"}

Verify the result

You can view the synthesized image by accessing the following domain name in your browser:

https://cri-zbtsehbrr8******-registry.oss-cn-shenzhen.aliyuncs.com/stylized-image.png

This domain name is used as an example. The actual domain name prevails.

Tutorial for object detection

You can use the object detection technology to construct a rectangular border for the desired object and track the object when multiple objects appear at the same time. Object detection applications are used to mark and identify a large number of different types of objects. In the following example, OpenCV Deep Neural Network (DNN) is used to perform multi-object detection.

Detection performance

The following table describes the source image of the object to be detected (left), and the detected objects by using OpenCV DNN (right). The right figure also displays the name of the detected objects and the detection accuracy.

Before you start

• For optimal user experience, join the DingTalk group (ID 11721331) and provide the following information:

  • The organization name, such as the name of your company.

  • The ID of your Alibaba Cloud account.

  • The region where you want to use GPU-accelerated instances, such as China (Shenzhen).

  • The contact information, such as your mobile number, email address, or DingTalk account.

• Perform the following operations in the region where the GPU-accelerated instances reside:

  • Create a Container Registry Enterprise Edition instance or Personal Edition instance. We recommend that you create an Enterprise Edition instance. For more information, see Step 1: Create a Container Registry Enterprise Edition instance.

  • Create a namespace and an image repository. For more information, see Step 2: Create a namespace and Step 3: Create an image repository.

• Install Serverless Devs and Docker
• Configure Serverless Devs

• Compile OpenCV.

  OpenCV must be compiled before you can use GPU acceleration. The following items describe how to compile OpenCV:

  • (Recommended) Use compiled OpenCV through Docker. Download address: opencv-cuda-docker and cuda-opencv.

  • Manually compile OpenCV. For more information, see the compilation guide.

• Upload the audio and video resources that you want to process to an Object Storage Service (OSS) bucket in the region where the GPU-accelerated instances are located. Make sure that you have the read and write permissions on the objects in the bucket. For more information about how to upload audio and video resources, see Upload objects. For more information about permissions, see Modify the ACL of a bucket.

Procedure

1. Create a project.

   s init devsapp/start-fc-custom-container-event-python3.9 -d fc-gpu-prj

   The following sample code shows the directory of the created project:

   fc-gpu-prj
   ├── code
   │   ├── app.py        # Function code.
   │   └── Dockerfile    # Dockerfile: The image Dockerfile that contains the code.
   ├── README.md
   └── s.yaml            # Project configurations, which specify how the image is deployed in Function Compute

2. Go to the project directory.

   cd fc-gpu-prj

3. Modify the configurations of files in the directory based on your business requirements.

   • Edit the s.yaml file.

     For more information about the parameters in the YAML file, see YAML specifications.

     edition: 1.0.0
     name: container-demo
     access: default
     vars:
       region: cn-shenzhen
     services:
       customContainer-demo:
         component: devsapp/fc
         props:
           region: ${vars.region}
           service:
             name: tgpu_object_detect_service
             internetAccess: true
           function:
             name: tgpu_object_detect_func
             description: test gpu for opencv
             handler: not-used
             timeout: 600
             caPort: 9000
             memorySize: 16384
             gpuMemorySize: 8192
             instanceType: fc.gpu.tesla.1
             runtime: custom-container
             customContainerConfig:
               #1. Make sure that the namespace:demo namespace and the repo:gpu-transcoding_s repository are created in advance in Alibaba Cloud Container Registry. 
               #2. Change the tag from v0.1 to v0.2 when you update the function and run s build && s deploy again. 
               image: registry.cn-shenzhen.aliyuncs.com/demo/gpu-object-detect_s:v0.1
             codeUri: ./code
           triggers:
             - name: httpTrigger
               type: http
               config:
                 authType: anonymous
                 methods:
                   - GET

   • Edit the app.py file.

     Example:

     # -*- coding: utf-8 -*-
     # python2 and python3
     
     from __future__ import print_function
     from http.server import HTTPServer, BaseHTTPRequestHandler
     import json
     import sys
     import logging
     import os
     import numpy as np
     import cv2
     import urllib.request
     
     class Resquest(BaseHTTPRequestHandler):
         def download(self, url, path):
             print("enter download:", url)
             f = urllib.request.urlopen(url)
             with open(path, "wb") as local_file:
                 local_file.write(f.read())
     
         def upload(self, url, path):
             print("enter upload:", url)
             headers = {
                 'Content-Type': 'application/octet-stream',
                 'Content-Length': os.stat(path).st_size,
             }
             req = urllib.request.Request(url, open(path, 'rb'), headers=headers, method='PUT')
             urllib.request.urlopen(req)
     
         def core(self):
             CLASSES = ["background", "aeroplane", "bicycle", "bird", "boat",
                    "bottle", "bus", "car", "cat", "chair", "cow", "diningtable",
                    "dog", "horse", "motorbike", "person", "pottedplant", "sheep",
                    "sofa", "train", "tvmonitor"]
             COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))
     
             print("[INFO] loading model...")
             prototxt = "/usr/src/app/m.prototxt.txt"
             model = "/usr/src/app/m.caffemodel"
             net = cv2.dnn.readNetFromCaffe(prototxt, model)
     
             msg = ""
             mode = ""
             if not cv2.cuda.getCudaEnabledDeviceCount():
                 msg = "No CUDA-capable device is detected |"
             else:
                 msg = "CUDA-capable device supported |"
                 net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
                 net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
     
             path = "/tmp/target.png"
             # Use the path of the OSS object in your Alibaba Cloud account. You must have the read and write permissions on the object. Read the image from your bucket. 
             self.download("https://your_public_oss/a.png", path)
             image = cv2.imread(path)
             (h, w) = image.shape[:2]
             blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843, (300, 300), 127.5)
     
             print("[INFO] computing object detections...")
             net.setInput(blob)
             detections = net.forward()
     
             # loop over the detections
             for i in np.arange(0, detections.shape[2]):
                 confidence = detections[0, 0, i, 2]
                 if confidence > 0.2:
                     idx = int(detections[0, 0, i, 1])
                     box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
                     (startX, startY, endX, endY) = box.astype("int")
                     cv2.rectangle(image, (startX, startY), (endX, endY), COLORS[idx], 2)
                     x = startX + 10 if startY - 15 < 15 else startX
                     y = startY - 15 if startY - 15 > 15 else startY + 20
                     label = "{}: {:.2f}%".format(CLASSES[idx], confidence * 100)
                     cv2.putText(image, label, (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)
                     print("[INFO] {}".format(label))
             cv2.imwrite(path, image)
     
             # Use the path of the OSS object in your Alibaba Cloud account. You must have the read and write permissions on the object. Read the image from your bucket. 
             self.upload("https://your_public_oss/target.jpg", path)
             msg = msg + " process image ok!"
     
             data = {'result': msg}
             self.send_response(200)
             self.send_header('Content-type', 'application/json')
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def pong(self):
             data = {"function":"object-detection"}
             self.send_response(200)
             self.send_header('Content-type', 'application/json')
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def dispatch(self):
             mode = self.headers.get('RUN-MODE')
     
             if mode == "ping":
                 self.pong()
             elif mode == "normal":
                 self.core()
             else:
                 self.pong()
     
         def do_GET(self):
             self.dispatch()
     
         def do_POST(self):
             self.dispatch()
     
     if __name__ == '__main__':
         host = ('0.0.0.0', 9000)
         server = HTTPServer(host, Resquest)
         print("Starting server, listen at: %s:%s" % host)
         server.serve_forever()

   • Edit the Dockerfile file.

     Example:

     FROM registry.cn-shanghai.aliyuncs.com/serverless_devs/opencv-cuda:cuda-10.2-opencv-4.2
     WORKDIR /usr/src/app
     RUN sed -i s@/archive.ubuntu.com/@/mirrors.aliyun.com/@g /etc/apt/sources.list
     RUN apt-get clean
     RUN apt-get update --fix-missing
     RUN apt-get install -y build-essential
     RUN apt-get install -y python3
     COPY . .
     CMD [ "python3", "-u", "/usr/src/app/app.py" ]
     EXPOSE 9000

4. Download the following files and store the files to the /code directory.

   • m.caffemodel

   • m.prototxt.txt

5. Build an image.

   s build --dockerfile ./code/Dockerfile

6. Deploy the code to Function Compute.

   s deploy

   Note

   If you run the preceding command repeatedly and service name and function name remain unchanged, run the use local command to use local configurations.

7. Configure provisioned instances.

   s provision put --target 1 --qualifier LATEST

8. Check whether the provisioned instances are ready.

   s provision put --target 1 --qualifier LATEST

   If the value of current is 1, the provisioned mode of GPU-accelerated instances is ready. Example:

   [2021-12-07 02:20:55] [INFO] [S-CLI] - Start ...
   [2021-12-07 02:20:55] [INFO] [FC] - Getting provision: tgpu_object_detect_service.LATEST/tgpu_object_detect_func
   customContainer-demo:
    serviceName:      tgpu_object_detect_service
    functionName:      tgpu_object_detect_func
    qualifier:       LATEST
    resource:        188077086902****#tgpu_object_detect_service#LATEST#tgpu_object_detect_func
    target:         1
    current:        1
    scheduledActions:    (empty array)
    targetTrackingPolicies: (empty array)

9. Invoke the function.

   • View online function versions

     s invoke
     FC Invoke Result:
     {"result": "CUDA-capable device supported | process image ok!"}

   • Detect objects

     s invoke -e '{"method":"GET","headers":{"RUN-MODE":"normal"}}'
     enter upload: https://your_public_oss/target.jpg # You can download this file to view the inference results.
     FC Invoke Result:
     {"result": "CUDA-capable device supported | process image ok!"}

10. Release the GPU-accelerated instances.

    s provision put --target 0 --qualifier LATEST

Verify the result

You can view the image after objects are detected by accessing the following domain name in your browser:

https://cri-zbtsehbrr8******-registry.oss-cn-shenzhen.aliyuncs.com/target2.jpg

This domain name is used as an example. The actual domain name prevails.