2023 Huawei Product Evaluation Officer - Voice of Developers + Huawei Cloud ModelArts Trial Experience Experience

2023 Huawei Product Evaluation Officer - Voice of Developers + Huawei Cloud ModelArts Trial Experience Experience

1. Activity introduction

Participate in HUAWEI CLOUD CodeArts, HUAWEI CLOUD ModelArts, Kunpeng DevKit, Ascend CANN**, **Sense MindSpore, 5 product experience tasks from the "user perspective". There are 3 experience methods for you to choose from. They are entry experience, sample code, and free experience.

As a one-stop AI development platform, Huawei ModelArts provides AI engineers with very powerful and comprehensive support, which can greatly improve the efficiency of AI model development.

2. Introduction to HUAWEI CLOUD ModelArts

• ModelArts is a one-stop development platform for AI developers. It helps you intelligently and efficiently create AI models and deploy them to the cloud, edge, and end with one-click model through AI development full-process management.

• AIGallery of ModelArts presets a large number of assets such as models, algorithms, data, and notebooks for beginners to use quickly; the automatic learning function of ModelArts can help users build AI models with zero code; ModelArts also provides a development environment for users You can write training codes in JupyterLab on the cloud or local IDE for AI model development.

• ModelArts Architecture
  

3. Introduction to AI Gallery

AI Gallery is an AI application market based on ModelArts. This function greatly reduces the threshold for AI model development. Based on AI Gallery's AI model market, even if users do not understand AI knowledge, they can directly use the model by subscribing to the model service. I think that with the advancement of technology and the development of society, AI must become more popular, and this function of AI Gallery is a great pioneer in this direction.

In this trial, taking the **"Supermarket Commodity Recognition"** model as an example, complete the free experience process from AI Gallery subscription model to Modelarts one-click deployment as an online service. The main steps are as follows:

Step 1: Subscribe to the model

Step 2: Deploy the online service using the subscription model

Click Online Deployment and configure parameters

After confirming the specification parameters, click "Submit" to start the deployment of online services. Wait for the service status to change to "Running", indicating that the service deployment is successful. Estimated duration is about 4 minutes

Step 3: Predict Outcomes

On the "Prediction" tab, click "Upload" to upload a test image, and click "Prediction" to view the prediction results

Step 4: Call the model through the API interface

API calls can be made through the HTTP/WS interface, and third-party programs can use the model more conveniently

4. JupyterLab of ModelArts

ModelArts supports opening notebooks online through the JupyterLab tool, and developing AI models based on PyTorch, TensorFlow, and MindSpore engines. The specific operation process is shown in the figure below.

You can find existing Notebook cases in AI Gallery, and click Run In ModelArts to enter Jupyter Lab for operation.

Five, trial experience

Through the in-depth experience of ModelArts, I have the following impressions

1. The product documentation and novice navigation are very friendly, and it is easy to get started

The product quick start navigation link is as follows: https://support.huaweicloud.com/qs-modelarts/modelarts_06_0006.html

2. The training framework is complete and compatible with mainstream open source codes

ModelArts Notebook has very practical functions, supports a variety of mainstream development languages, and has a large number of commonly used toolkits built in. It can write codes and run programs online, which greatly improves development efficiency. At the same time, Notebook can directly access platform data sets, training tasks and other resources to achieve seamless connection between code and resources.

3. It is very convenient to deploy the model

Just click deploy to generate online prediction service. AI prediction capabilities can be provided externally in real time by calling the service interface. The platform will flexibly adjust computing resources according to usage conditions to ensure stable service performance.

4. Preset algorithm recurrence example, you can get started quickly

ModelArts AI Gallery provides a wealth of algorithm reproduction examples, all of which are verified and reliable codes, including computer vision, natural language processing and other fields, which lowers the threshold for algorithm implementation and allows you to get started quickly.

5. Model management is versioned, supporting one-click deployment and online

After training, the model can be published as an online service for deployment, and version management is supported. The deployment of the new version of the model can be completed with one click, without starting a new service. It supports cloud deployment and edge deployment at the same time, which is very in line with actual product needs.

5. Real-time monitoring of training logs, clear resource statistics

ModelArts provides a complete log system and system monitoring. You can clearly observe the logs of each training job, including loss function curves and data of various indicators, and display the GPU, CPU, and memory usage of each node in real time. It can greatly improve the efficiency of model debugging.

6. High flexibility in storage and computing resources

According to the needs of the project, you can choose the appropriate configuration of OBS storage space, as well as scalable GPU and CPU resources. Resources can be released and expanded on demand, truly realizing pay-as-you-go

Through the trial of ModelArts this time, it can be seen that ModelArts has industry-leading advantages in terms of ease of use, complete functions, and elastic resource scalability. It is an excellent AI development and production platform.