Learn Before
Augmented Input Formula for External Memories
When utilizing external memories or datastores for language modeling, the retrieved information is integrated with the original input to form an augmented input for the Large Language Model (LLM). If the original input is denoted as and the retrieved context elements are , they are combined using a function or template . The resulting augmented input, , is mathematically expressed as: . The accompanying illustration demonstrates how these external memories are integrated into the language modeling workflow.
0
1
Tags
Foundations of Large Language Models
Ch.2 Generative Models - Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
Using Retrieved Context to Improve Attention
Retrieval-Based Methods as a Solution for Long-Context Processing
Unsuitability of External Memory for Streaming Contexts
k-NN as a Popular Retrieval-Based External Memory Method
Computational Cost of External Memory Models
Architectural Design for a Real-Time Chat Application
A company is building a question-answering system to help employees query a massive, static knowledge base of over 100,000 internal documents. The core language model has a fixed input size that is much smaller than the total size of the knowledge base. Which approach is the most effective and scalable for ensuring the model can access the necessary information to answer specific user queries accurately?
Evaluating the Use of External Memory Systems for LLMs
Augmented Input Formula for External Memories
Comparison of External Memories in LLMs