Praise Be! Being an AI Evangelist.

Becoming an AI Evangelist // Alex Volkov // MLOps Podcast #215

While it's not quite dreaming of electric sheep, Alex did talk about one case of multimodality where the inputs from an EEG and fMRI signals to invoke a specific state in your brain for lucid dreaming, which is mind-blowing.

So yeah, we talked about multimodality, speed challenges, evaluating and fine-tuning embedding models, user experience, and more, all feeding into his AI evangelism! We also talked about his journey seeing the light, from leaving management roles to podcast host, with a Denver MLOps Community meetup thrown in!

Below is an excellent example of why you should check out the upcoming IRL meet-ups!

Video || Spotify || Apple

We're thrilled to invite you to our first virtual conference, Infer: Spring '24.
This conference, which will take place on March 20 between 11:00 AM and 3:30 PM EST, will bring together thought leaders and enthusiasts in ML and AI to discuss the hottest topics shaping the industry today, the challenges practitioners face, and best practices for overcoming them.

Our line-up brings years of experience from leading companies such as JFrog, Uber, Google, and more.

Engaging talks, Q&As, and workshops covering:

Key Principles for Running LLMs in Production
Dealing with Hallucinations
All You Need to Know About LLM Gateways
The Life of a Feature - a Journey through Space and Time

Plus - a special announcement!
We'll be unveiling an exciting new Qwak feature live during the conference!

Check out the full agenda and save your spot below.

Join us at Infer

Managing Data for Effective GenAI Application // Anu Arora & Anass Bensrhir // MLOps Podcast #216

They say things are as easy as ABC. Do you know why they stop there? Because DE is hard. Data Engineering is challenging even with the text-to-SQL models that have come out.

We discussed all the challenges they encounter when GenAI use cases go from MVP to production, how ETL is now a little bit different, and, of course, data quality. Building on the data quality point, we also discussed keeping things updated in a use case like a RAG model to ensure that the most up-to-date information is used in your RAG.

So, data engineering is still hard, but watching is as easy as clicking the link below!

Video || Spotify || Apple

Thanks to the kind folks at the Linux Foundation, we can offer 15% off passes* for KubeCon + CloudNativeCon Europe on March 19-22 in Paris!
That's 15% off access to the AI hub and talks about Security, Observability, Platform Engineering, Service Mesh, and more, like:

Platform Building Blocks: How to Build ML Infrastructure with CNCF Projects - Yuzhui Liu, Team Lead & Leon Zhou, Software Engineer, Bloomberg.
Accelerate Kubernetes Data-Intensive Apps with CloudNative Local Storage & Distribute Cache System - Simon YN Zhao & Zhou Mingming, DaoCloud.
Strategies for Efficient LLM Deployments in Any Cluster - Angel M De Miguel Meana, VMware & Francisco Cabrera, Microsoft

To get your discount and access to 4 days of 100+ sessions, enter the code MLOPS_KUBECON at the checkout.
*Ts&Cs apply: 15% off applies to All Access or KubeCon + CloudNativeCon passes (corporate or individual). Entrance only, travel and accommodation not included.

💡Job of the week

Senior AI Developer // Desia (UK/Europe, hybrid/remote)
Desia AI is an early-stage, London Based pre-seed start-up with focus on Financial knowledge management, building a system of intelligence empowered by AI-driven automation and analytics.
Responsibilities:

Leading the development of new generative AI features and addressing complex problems, such as multi-hop question answering and interacting with symbolic systems.
Building Semantic Knowledge Graph along with other engineers that powers the unique value proposition.
Working within a small, dynamic team that promotes maximum autonomy and encourages rapid innovation.

Requirements:

Experienced with foundational Language Models (SLM / LLMs) in production environments.
Skilled in deploying machine learning solutions to production.
Good understanding of data modelling for search/retrieval and reranking.
Strong NLP skills, especially in Name Entity Disambiguation.

From downtown...!
I got flashbacks to NBA Jam knockout tournaments on the SNES when I heard this idea. Inspired by the NCAA Basketball's March Madness, we've partnered with the folks at Seaplane to create a competition to crown one model the All-Star!

He's heating up...!
With battles starting on March 25th, March Model Madness will pit 16 of the top FM and LLMs against each other in a competitive battle measured on Chat, Instruct, and Code. MMM will be a blind competition where the winner, remaining unexposed, will proceed to the next competitive round while only the loser will be exposed. Sounds fun, right?

He's on fire!
Well, it gets better! You can get involved now by submitting and upvoting proposed prompts to be used in the battle. Then, when the battle phase starts, you get to vote for the best results. There are prizes to be won for participating, including AirPods Max, Amazon Gift Cards, Seaplane swag, and MLOps Community swag.

Start submitting your prompts now!

PIXART-α: A Diffusion Transformer Model for Text-to-Image Generation

If a picture paints a thousand words, does an art gallery count as an LLM?

This blog doesn't answer that but does look at PIXART-α, a diffusion transformer model designed for text-to-image generation. It looks at the architecture and training strategy for Pixart-α, running it using HuggingFace Diffusers and managing experiments using Weights & Biases. The blog also compares the quality of images generated with Stable Diffusion XL.

With thanks to Soumik Rakshi for their contribution.