Training Data & Token Window: A Comprehensive Guide

In this informative and comprehensive video, we delve into the intricacies of training Ask Sage with your data, and the concept of a Token Window (or Context Window). We start by explaining the three types of memory and their significance in the context of Ask Sage.

We then move on to discuss the agnosticism of Ask Sage towards various Large Language Models (LLMs), highlighting its versatility and adaptability. The video further explains the difference between fine-tuning and embeddings, providing a clear understanding of these crucial aspects of machine learning.

The concept of vectors and how vector search operates based on the context window is also covered in detail. We then demonstrate how embeddings are injected into Ask Sage, enhancing its capabilities and performance.

The video also provides a deep dive into the Token/Context Window, explaining its importance and how it influences the performance of the model. We then discuss how to manage datasets and train data effectively, ensuring optimal results.

We also showcase our File Ingest plugin, which allows for the easy ingestion of various file types including PDF, Word, Excel, PowerPoint, and more. This is followed by a concrete example of a Request for Proposal (RFP) response generation, demonstrating the practical application of the concepts discussed.

The video concludes with a demonstration of our new Search Dataset Plugin, which streamlines the down-selection of your relevant data, and a comparison of different models. We also provide tips on creating good and bad prompts, and share more examples to provide a better understanding.

This video serves as a comprehensive guide for anyone looking to understand and utilize Ask Sage to its full potential.

Timestamps:
00:00 Introduction
00:22 3 types of memory
02:32 Ask Sage LLM Agnosticism
03:13 Fine tuning
04:20 Embeddings
04:38 Vectors & Vector database
06:37 Embeddings injection in Ask Sage
07:51 Token / Context Window
15:24 Datasets & Training
22:45 File Ingest Plugin
26:14 RFP response example
28:23 Search Dataset Plugin
34:14 Model tokens window
36:45 Good and Bad Prompts
38:39 More examples
41:11 Model comparison
42:14 Conclusion