Scaling Language Model Tuning with Snorkel AI - A Case Study

Explore a 22-minute case study on how Snorkel AI scales language model tuning. Discover how research scientist Chris Glaze and his team enhanced the efficiency and scalability of fine-tuning Large Language Models (LLMs) using machine learning techniques. Learn about their innovative approach using programmatic labeling on Snorkel Flow to develop two guiding models: one for categorizing instructions and another for assessing response quality. Understand how these models helped curate 20,000 prompt-response pairs down to the most effective 10,000 for fine-tuning the RedPajama LLM. Gain insights into their fine-tuned version, which outperformed the baseline model in human evaluations across all measured categories. Delve into topics such as golden data requirements, CLE's use of large language models, instruction classification, response quality modeling, and experimental win rates. Enhance your understanding of cutting-edge techniques in AI model tuning and machine learning.

Introduction
Overview of large language model development
How finetuning requires golden data
How CLE uses large language models
Guiding models
Case study
Instruction classification
Instruction classes
Response quality model
Win rates
Experiment
Summary
Outro