Amazon SageMaker JumpStart provides built-in algorithms, pre-trained models, and pre-built solution templates to assist data scientists and machine learning practitioners in quickly training and deploying ML models. This post looks at how to use the text classification and fill-mask models on Hugging Face with SageMaker JumpStart for text classification on a custom dataset. The tutorial also demonstrates how to perform real-time and batch inference for these models.
JumpStart’s text classification with Hugging Face provides transfer learning on all pre-trained models available on Hugging Face. Depending on the number of class labels in the training data, a classification layer is added to the pre-trained Hugging Face model. Training can be done even with a smaller dataset.
The post also outlines how to set up the SageMaker execution role and the necessary setup steps before running the notebook, such as installing SageMaker and importing necessary libraries. It then goes into running inference on the pre-trained model, fine-tuning the pre-trained model on a custom dataset, and performing batch inference with the Hugging Face text classification algorithm.
The tutorial also covers the prerequisites needed to follow the guide, including installing packages, setting up the SageMaker execution role, and importing necessary libraries. It moves on to show how to run inference on the pre-trained model and fine-tune it with a custom dataset. Moreover, it shows how to train using a custom training dataset, with demonstrations on coding for the entire process.
Additionally, it shows how to fine-tune any Hugging Face fill-mask or text classification model and perform batch inference with the Hugging Face text classification algorithm. SageMaker JumpStart also supports fine-tuning of any pre-trained fill-mask or text classification Hugging Face model.
Batch inference is particularly useful when generating predictions from a trained model on a large dataset where latency isn’t a concern. It’s helpful in preprocessing datasets to remove noise or bias, running inference from large datasets, running inference when a persistent endpoint isn’t needed, and associating input records with inferences.
In the end, the authors conclude by summarizing the features of SageMaker Hugging Face text classification algorithm, giving users the ability to perform transfer learning on a custom dataset using a pre-trained model, and run inference on sizable datasets via batch inference.
