DevEnviron
It is challenging to set up a development environment, select an AI algorithm framework and algorithm, debug code, install software, and accelerate hardware. To help users overcome these challenges, ModelArts simplifies the entire development process and lowers the development threshold. Figure 1 shows the algorithm development process.
- Support for all popular AI algorithm frameworks
In the machine learning and deep learning fields, popular open source training and inference frameworks include TensorFlow, PyTorch, MXNet, and MindSpore. ModelArts supports all popular AI computing frameworks and provides a user-friendly development and debugging environment. It supports traditional machine learning algorithms, such as logistic regression, decision tree, and clustering, as well as multiple types of deep learning algorithms, such as the convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory (LSTM).
- Simplified algorithm development for distributed training
Deep learning usually requires large-scale GPU clusters for distributed acceleration. For existing open source frameworks, algorithm developers need to write a large amount of code to implement distributed training on different hardware, and the acceleration code varies depending on the framework. The MoXing framework of ModelArts, which is a lightweight distributed framework built on deep learning engines such as TensorFlow, PyTorch, MXNet, and MindSpore, addresses these pain points. In addition, the distributed computing engine features higher performance and is easier to use. Figure 2 shows the code developed by a developer based on MoXing.
- Simplified parameter tuning: Multiple parameter tuning skill packages are integrated, for example, the data augmentation policy, which simplifies model tuning for AI algorithm engineers.
- Simplified distributed acceleration: Automatic distributed acceleration of standalone coding simplifies distributed acceleration and improves performance, eliminating the need for algorithm engineers to have knowledge of distribution.
Last Article: Data Management
Next Article: Model Training
Did this article solve your problem?
Thank you for your score！Your feedback would help us improve the website.