During its AWS re:Invent event today, AWS announced several updates to Amazon SageMaker, which is a platform for building, training, and deploying machine learning models.
It introduced new features that are designed to improve the model deployment experience, including the introduction of new classes in the SageMaker Python SDK: ModelBuilder and SchemaBuilder.
ModelBuilder, selects a compatible SageMaker container to deploy to and captures the needed dependencies. SchemaBuilder manages the serialization and deserialization tasks of inputs and outputs from the models.
RELATED CONTENT:
“You can use the tools to deploy the model in your local development environment to experiment with it, fix any runtime errors, and when ready, transition from local testing to deploy the model on SageMaker with a single line of code,” Antje Barth, principal developer advocate at AWS, wrote in a blog post.
SageMaker Studio was also updated with new workflows for deployment, which provide guidance to help choose the most optimal endpoint configuration.
SageMaker was also updated with new inference capabilities, which helps reduce deployment costs and latency. The new inference capabilities allow you to deploy one or more foundation models on a single endpoint and control the memory and number of accelerators assigned to them.
It also monitors inference requests and automatically routes them based on which instances are available. According to Amazon, this new capability can help reduce deployment costs by up to 50% and reduce latency by up to 20%.
There were also a few updates within Amazon SageMaker Canvas, which is a no-code interface for building machine learning models. Natural language prompts can now be used when preparing data.
In the chat interface, the application provides a number of guided prompts related to the database you are working with, or you can come up with your own. For example, you can ask it to prepare a data quality report, remove rows based on certain criteria, and more.
In addition, you can now use foundation models from Amazon Bedrock and Amazon SageMaker Jumpstart. According to the company, this new capability will enable companies to deploy models that are designed for their unique business.
SageMaker Canvas handles all the training and allows you to fine-tune the model once it is created. It also provides analysis of the created model and displays metrics like perplexity and loss curves, training loss, and validation loss.
