Uređivanje

Dijeli putem

Model fine-tuning concepts

Fine-tuning helps you adapt pre-trained AI models to work better with your specific data and use cases. This technique can improve model performance while requiring less training data than building a model from scratch.

This article covers:

  • What fine-tuning is and how it works
  • When to use fine-tuning versus other approaches
  • How to select and prepare models for fine-tuning
  • Best practices for iterating and improving your results

Prerequisites

Before you start, you should have:

  • Basic understanding of machine learning concepts
  • Familiarity with your specific use case and data requirements
  • Access to sample data for training and validation

What is fine-tuning?

Fine-tuning is a machine learning technique that adapts a pre-trained model to perform better on your specific task. Instead of training a model from scratch, you start with a model that already understands general patterns and adjust it to work with your data.

This approach leverages transfer learning—using knowledge gained from one task to improve performance on a related task. Fine-tuning is particularly effective when you have limited training data or want to build on existing model capabilities.

When to fine-tune

Fine-tuning works well when you have a small amount of data and want to improve your model's performance. By starting with a pre-trained model, you can use the knowledge the model already learned and adjust it to better fit your data. This approach helps you improve your model's performance and reduces the amount of data needed for training.

When you have a large amount of data, you usually don't need to fine-tune your model. You can train your model from scratch and get good results without fine-tuning. However, fine-tuning can still help if you want to boost your model's performance. You might also want to fine-tune your model if you have a specific task that's different from the task the pre-trained model was originally trained on.

You might avoid costly fine-tuning by using prompt engineering or prompt chaining. These techniques help you generate high-quality text without fine-tuning.

Select a pre-trained model

Choose a pre-trained model that fits your task requirements. Many pre-trained models are available, each trained on different tasks. Pick a model trained on a task similar to yours. This choice helps you use the knowledge the model already learned and adjust it to fit your data.

HuggingFace models are a good place to start when looking for pre-trained models. The HuggingFace models are grouped into categories based on the task they were trained on, making it easy to find a model that fits your task.

These categories include:

  • Multimodal
  • Computer vision
  • Natural language processing
  • Audio
  • Tabular
  • Reinforcement learning

Check if the model works with your environment and tools. For example, if you're using Visual Studio Code (VS Code), you can use the Azure Machine Learning extension for VS Code to fine-tune your model.

Check the status and license of the model. Some pre-trained models are available under an open-source license, while others require a commercial or personal license. All the models on HuggingFace include license information. Make sure you have the necessary permissions to use the model before fine-tuning it.

Prepare your sample data

Preparing your sample data involves cleaning and preprocessing your data to make it suitable for training. You should also split your data into training and validation sets to evaluate the performance of your model. The format of your data should match the format expected by the pre-trained model you use. You can find this information with the models on HuggingFace in the Instruction format section of the model card. Most model cards include a template for building a prompt for the model and some pseudocode to help you get started.

Iterate on your model

After you fine-tune your model, evaluate its performance on the validation set. Use metrics such as accuracy, precision, recall, and F1 score to evaluate the performance of your model. If the performance of your model isn't satisfactory, adjust the hyperparameters, change the architecture, or fine-tune the model on more data. You can also examine the quality and diversity of your data to see if there are any issues that need to be addressed. As a general rule, a smaller set of high quality data is more valuable than a larger set of low quality data.

See also

To learn more about fine-tuning AI models, check out the following resources:

When utilizing AI features, review: Developing Responsible Generative AI Applications and Features on Windows.

Related content

  • Last updated on