AlexaTM 20B model is now available in SageMaker JumpStart

Today, we announce the public availability of Amazon’s state-of-the-art Alexa Teacher Model 20 Billion parameter model (AlexaTM 20B) in SageMaker JumpStart. Customers can access the AlexaTM 20B model programmatically to run inference using APIs available in SageMaker Python SDK.

The Alexa Teacher Model (AlexaTM) program by Amazon Alexa AI is designed to build large-scale, multi-task, multi-lingual Deep Learning (primarily Transformer-based) models aiming to improve generalization without requiring large amount of data from downstream tasks. Leveraging large-scale pre-training, teacher models can generalize well to learn new tasks from sparse data and help developers improve accuracy on downstream tasks. AlexaTM 20B is Alexa AI’s largest model in size, and it has shown competitive performance on common NLP tasks and benchmarks (SuperGLUE and XNLI).

Amazon SageMaker JumpStart is the Machine Learning (ML) hub of SageMaker that offers 350+ built-in algorithms, pre-trained models, and pre-built solution templates to help customers get started with ML fast. Pre-trained models hosted in JumpStart include publicly available State-of-the-Art (SOTA) models from popular model hubs such as TensorFlow, PyTorch, Hugging Face and MXNet, and support popular ML tasks such as object detection, text classification, and text generation. To help data scientists and ML practitioners get started quickly and securely, contents are stored in an AWS repository and come with training and inferencing scripts compatible with SageMaker features. Customers can fine-tune models using their own data or deploy as-is for inferencing.

AlexaMT 20B can be used in all regions where Amazon SageMaker is available.

Visit Alexa Science blog to learn more about the model and AlexaTM in JumpStart launch blog to learn more about how to access it using JumpStart. To browse all models available in SageMaker JumpStart, please visit SageMaker JumpStart ML Hub.