Tutorial: Fine-Tuning the Reader Model in SDK

Learn how to improve the performance of your pipeline by fine-tuning the Reader model. This tutorial also teaches you how to store the fine-tuned model in Hugging Face and how to update your pipeline to use it.

  • Level: Advanced
  • Time to complete: 30 minutes
  • Prerequisites:
    • This tutorial assumes good knowledge of NLP and Python.
    • You must be an Admin to complete this tutorial.
    • More detailed prerequisites are listed in the Jupyter Notebook we prepared for you.
  • Goal: After completing this tutorial, you will have a question-answering pipeline with a Reader using a model you have fine-tuned.

You will complete this tutorial in Jupyter Notebooks that you'll access from deepset Cloud. This way, you'll be able to immediately try out the code and see what it does. We prepared a notebook that guides you through all the steps.

To access the tutorial:

  1. Go to deepset Cloud and click Notebooks.
  2. Select GPU and wait until the server is ready.
  3. Once the server is created, click Go to JupyterLab.
  4. From the examples folder, open a notebook called 03_reader_finetuning.ipynb and follow the instructions there.