---
title: "Lab 1 - Few Shot Learning Using SageMaker"
weight: 11
---

## Introduction
In this Lab, we'll gain a introduction to SageMaker through a practical use case of Few-Shot Learning using a large language model. Specifically, we'll utilize the 🤗 [SetFit](https://github.com/huggingface/setfit) library which implements a few-shot learning [algorithm](https://arxiv.org/abs/2209.11055) of the same name to train a model using [SageMaker Training](https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works-training.html) and then deploy it to a [SageMaker Endpoint](https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works-deployment.html) for inference. 

## Background and Use Case
We will use the [Banking 77](https://github.com/PolyAI-LDN/task-specific-datasets/tree/master/banking_data) dataset which classifies customer online banking queries into one of 77 intents. Few examples of the dataset are shown below. 
- What can I do if my card still hasn't arrived after 2 weeks? -> card_arrival
- How do you decide what the exchange rates are? -> exchange_rate
- Please let me know how to change my PIN? -> change_pin

While the dataset contains over 10,000 we will only use a small fraction of it to train the model and still achieve great results. This is enabled by the few-shot learning SetFit algorithm which is able to learn from a small number of examples. The algorithm is illustrated below [source](https://github.com/huggingface/setfit) 

![SetFit](../assets/images/lab1/setfit.png)

This algorithm works by first generating pairs of matching and non-matching category examples from the training data. It then fine-tunes a SentenceTransformer model on these pairs to distinguish between matching and non-matching examples, similar to a binary classification but using a contrastive loss function such as Cosine Similarity Loss or Triplets Loss. The outcome of this training is that the SentenceTransformer model is able to encode the input text into a vector space where similar examples are close to each other and dissimilar examples are far apart. We can use these encoded values as inputs to train a classifier on top of the SentenceTransformer model which by default is a Logistic Regression model from SciKit Learn.

We will launch a SageMaker Training job to train the model, track the experiment using SageMaker Experiments, and then deploy the trained model to a SageMaker Endpoint for inference.

## Instructions
To get started, make sure to first complete all the [workshop setup](../../1-introduction/4-workshop-set-up) steps. Once you are in the SageMaker Studio root directory, double-click the `lab1` directory to navigate inside then double-click the `few_shot_learning.ipynb` file to open the Lab1 notebook. If prompted to select a Kernel as per the popup below, then select the following options **Image**: Data Science; **Kernel**: Python 3; **Instance Type**: ml.t3.medium

![KernelSelect](../assets/images/lab1/kernel_select.JPG)

Confirm that your notebook shows the instance and kernel selection that is consistent with the screenshot below. 

![KernelSelect](../assets/images/lab1/notebook_picture.JPG)

You are now ready to run the notebook. The notebook allows you to run code snippets interactively by either clicking the play icon in the toolbar or pressing `Shift+Enter` on your keyboard.  

## Conclusion
In this lab we saw how we can train a model using SageMaker Training and deploy it to a SageMaker Endpoint for inference. We also saw how we can use SageMaker Experiments to track the experiment.