September 27, 2023 | Meghana Denduluri

This post gives you a quick overview of the key differences between RAG and Fine-tuning in LLMs

TL ; DR

• Retrieval Augmented Generation: For providing an external context
  • Inserting additional data into the prompt of a model at inference time using traditional databases, vector databases, API calls, etc.
• Fine Tuning: For adjusting a model's behavior
  • Modifying the weights of a pre-trained model, using labeled domain- and task-specific data

Table of Contents

Retrieval-augmented generation ( RAG ) :

To enhance the quality and relevance of the generated text, Retrieval-augmented generation is a method in natural language processing that combines the strengths of retrieval-based models and generative models.

To grasp the concept of RAG, it's essential to comprehend its two primary components: retrieval models and generative models.

Retrieval models:

Generative models:

<aside> ✅ RAG marries these two approaches to mitigate their individual shortcomings.

</aside>

<aside> ✅ A retrieval-based model sources relevant data from a knowledge base or document collection based on a query or context.

</aside>

<aside> ✅ This extracted data then serves as input or supplementary context for the generative model.

</aside>

<aside> ✅ This fusion allows the generative model to tap into the precision and specificity of the retrieval-based model, resulting in more relevant and accurate text.

</aside>

Fine-Tuning :

Fine-tuning is a process where a pre-trained model is further trained on a specific task or dataset to improve its performance on that task. It allows the model to learn the specific patterns and relationships that are relevant to the target task.