Super Data Science: ML & AI Podcast with Jon Krohn

801: Merged LLMs Are Smaller And More Capable, with Arcee AI's Mark McQuade and Charles Goddard

Jul 16, 2024

Mark McQuade and Charles Goddard from Arcee AI discuss merging LLMs efficiently, using MergeKit and evolutionary algorithms. They explore commercial applications, compare MoE vs. MoA, and highlight the advantages of smaller language models. The podcast also covers the Spectrum Project for efficient training and the future of SLMs.

01:17:05

Episode guests

Mark McQuade

Charles Goddard

AI Summary

Highlights

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

Model merging allows for combining multiple LLMs without increasing size, enhancing efficiency.

Evolutionary model merging optimizes performance parameters, demonstrating cost efficiencies and domain-specific capabilities.

Utilizing specific modules for training in Spectrum accelerates model training by 40-50%, optimizing performance.

Deep dives

Model Merging and its Advantages in AI

Model merging, a technique discussed in the podcast, involves combining pre-trained weights of neural networks into a single network that captures the strengths of multiple networks. This technique allows for superior performance without increasing model size, making it cost-effective and efficient for training various language models efficiently. By targeting specific modules of the network to train while freezing others, model merging eliminates the need to train language models from scratch each time, offering significant advantages in terms of cost savings and improved performance.

Efficient Model Merging for Cost and Speed Optimization

01:37

Automated Model Merging for Enhanced Results

01:48

Data is the New Gold

01:09

Efficiency in Model Activation

01:23

Intro

4min

Exploring Model Merging for Enhanced Neural Networks

12min

MergeKit: Empowering Organizations with Custom Language Models and Evolutionary Model Merging

7min

Innovative Model Merging Using Evolutionary Algorithms

7min

Benefits of Model Merging and Case Studies on Smaller Merged Models

4min

Value of Merged Large Language Models in the Enterprise

7min

Efficient Training of Language Models with Sparse Upcycling and MergeKit

23min

Evolution of Models: Small and Powerful

9min

Enhancing Language Models Through Model Merging and Innovative Training Approaches

4min

Merged LLMs are the future, and we’re exploring how with Mark McQuade and Charles Goddard from Arcee AI on this episode with Jon Krohn. Learn how to combine multiple LLMs without adding bulk, train more efficiently, and dive into different expert approaches. Discover how smaller models can outperform larger ones and leverage open-source projects for big enterprise wins. This episode is packed with must-know insights for data scientists and ML engineers. Don’t miss out!

Interested in sponsoring a SuperDataScience Podcast episode? Email natalie@superdatascience.com for sponsorship information.

In this episode you will learn:

• Explanation of Charles' job title: Chief of Frontier Research [03:31]

• Model Merging Technology combining multiple LLMs without increasing size [04:43]

• Using MergeKit for model merging [14:49]

• Evolutionary Model Merging using evolutionary algorithms [22:55]

• Commercial applications and success stories [28:10]

• Comparison of Mixture of Experts (MoE) vs. Mixture of Agents [37:57]

• Spectrum Project for efficient training by targeting specific modules [54:28]

• Future of Small Language Models (SLMs) and their advantages [01:01:22]

Additional materials: www.superdatascience.com/801

Get the Snipd
podcast app

Unlock the knowledge in podcasts with the podcast player of the future.

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Save any
moment

Hear something you like? Tap your headphones to save it with AI-generated key takeaways

Share
& Export

Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Super Data Science: ML & AI Podcast with Jon Krohn

801: Merged LLMs Are Smaller And More Capable, with Arcee AI's Mark McQuade and Charles Goddard

Episode guests