Meta’s Chameleon, RAG with Autoencoder-Transformed Embeddings, and more #30

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

Good morning, AI enthusiasts! This week we are diving into some interesting discussions on transformers, BERT, and RAG, along with some interesting collaboration opportunities for building a bot, a productivity app, and more.

What’s AI Weekly

This week in What’s AI, I dive into how multimodal models actually work. Thanks to Chameleon, Meta’s open-source alternative to multimodal models, which has very useful details for building such a powerful model. I also talk about where exactly this is useful and how it differs from other models, like GPT-4 or Llama. Read the complete article here, and if you prefer a video, watch it here!

— Louis-François Bouchard, Towards AI Co-founder & Head of Community

Learn AI Together Community section!

Featured Community post from the Discord

Eschnou just released OpenGPA, an open-source General Purpose Agent. It is like a self-hosted, customizable version of ChatGPT that you can extend with custom actions to leverage your enterprise data and APIs. You can run this with open models as well as popular commercial LLMs. Check it out here and support a fellow community member. Share your feedback and requirements for Agentic in an enterprise context in the thread!

AI poll of the week!

The results are surprising, but not at the same time. Are factors other than accuracy (like price, speed, etc.) also guiding the decision? Tell us in the Discord thread!

Collaboration Opportunities

The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel! Keep an eye on this section, too — we share cool opportunities every week!

1. Gere030199 has built an AI Discord bot and needs help with the last leg of the development process. If you are good with Python, AI, ML, APIs, py-cord, or setting up a machine/server, connect with him in the Discord thread!

2. Sara.v is looking for someone interested in working on a gamified productivity app. If you have some experience with game design and app development, reach out in the thread!

3. Baadror is starting a hands-on LLM learning journey and looking for partners. If this sounds like something you would be interested in, contact him in the thread!

Meme of the week!

Meme shared by ghost_in_the_machine

TAI Curated section

Article of the week

BERT: In-depth exploration of Architecture, Workflow, Code, and Mathematical Foundations by Jaiganesan

If you’ve been in the AI field for a while, you’ve likely come across BERT multiple times. Introduced in 2018, BERT has been a topic of interest for many, with many articles and YouTube videos attempting to break it down. However, this article takes a different approach by delving into Embeddings, Masked Language Model Tasks, Attention Mechanisms, and Feed-Forward Networks.

Our must-read articles

1. A Novel Retrieval-Augmented Generation with Autoencoder-Transformed Embeddings by Shenggang Li

It’s common to use direct RAG methods like the shortest cosine distance retriever. However, these methods can result in irrelevant prompts due to noise in the knowledge base. By the end of this post, you’ll understand how to use RAG with Autoencoder-Transformed Embeddings, a method proposed here. The author also includes experimental data, mathematical background, and proofs to support this approach.

2. Want to Learn Quantization in The Large Language Model? By Milan Tamang

Quantization is a method of compressing a larger size model (LLM or any deep learning model) to a smaller size. In this article, you’ll learn about the what and why of quantization. Next, you’ll dive in further to understand the how of quantization with some simple mathematical derivations. Finally, we’ll write some code together in PyTorch to perform quantization and de-quantization of LLM weight parameters.

3. Understanding Mamba and Selective State Space Models (SSMs) by Matthew Gunton

The Transformer architecture has been the foundation of most major large language models (LLMs) on the market today, delivering impressive performance and revolutionizing the field. In this blog, we’ll explore a novel block architecture that aims to achieve the power of LLMs without the scalability limitations of traditional Transformers.

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI