Model Development

ModelArts Studio provides model development functions. The platform supports full-process model lifecycle management, ensuring efficient and accurate execution of every phase from data preparation to model deployment, providing powerful intelligent support for real-world applications.

• Model evaluation: The platform provides comprehensive model evaluation functions to ensure the effectiveness of models in real-world applications. The automatic evaluation function allows you to continuously monitor key metrics such as model precision and recall rate during training to promptly identify and address potential issues. Model evaluation functions help you verify the accuracy and reliability of models in diverse application scenarios. Additionally, the platform supports rule-based automatic evaluation. For NLP models, it displays metrics such as accuracy, F1 score, BLEU, and ROUGE. It also supports manual evaluation and customization of evaluation metrics. Model performance can be scored based on different evaluation metrics on the manual evaluation page.
• Model compression: Before model deployment, compression is a key step to improve inference performance. Through model compression, the memory occupied in the inference process can be effectively reduced, thereby saving resources and improving the computing speed.
• Model deployment: The platform provides a one-click deployment feature, allowing you to easily deploy trained models to either a cloud or on-premises environment. It supports multiple deployment options to meet the requirements of different scenarios. With flexible APIs, models can be seamlessly integrated into various applications.
• Model calling: After a model is deployed, you can use the model calling function to quickly access the model's services. The platform provides efficient APIs to help you easily integrate models into your applications and implement functions such as intelligent dialogue and text generation.