LLM fine-tuning and model selection + other resources
In the new year, we're?bringing fresh topics and articles.?This month,?we're taking a closer look at LLM tuning, playing?around with PyMC & Arviz, and sharing knowledge?on building ML systems using feature stores.
Happy reading!
---
Case studies & practical MLOps
>?How to Build Machine Learning Systems With a Feature Store?- First of all, an?introduction to a unified architecture for ML systems.?This architecture is built around the concept of FTI pipelines, with a feature store as the central component. Jim Dowling also presents here how this architecture applies to different classes of ML systems, discusses MLOps and testing aspects, and looks at some example implementations.
> Mikiko Bazeley: What I Learned Building the ML Platform at Mailchimp - Secondly, an article that is a story told by ?????? Mikiko B. about her journey and the valuable lessons she learned while building the ML platform. Mikiko shares her?insights on the evolution, challenges, and objectives of Mailchimp's ML platform, emphasizing the importance of feedback and her key learnings from the process.
---
Guides & tutorials
>?LLM Fine-Tuning and Model Selection Using Neptune and Transformers?- Now, let's have a look at the world of Large Language Models. In this article, Pedro Gabriel Gengo Louren?o talks about insights and strategies for selecting the best LLM model and conducting efficient fine-tuning (even when resources are constrained). Ha also covers how to reduce a model’s memory footprint, speed up training, and share best practices for monitoring. ?
>?Logging PyMC and Arviz Artifacts on Neptune?-?And lastly, a guide ideal if you're interested in learning Bayesian modeling, especially in situations with limited or uncertain data. You'll learn how to integrate neptune.ai into your Bayesian workflow, using PyMC and ArviZ for modeling and visualization, as well as?simplifying the management of complex artifacts.? ?
---
领英推荐
Tools
---
ML Platform Podcast [extras]
> There are some new extras on the Neptune YouTube channel. In just a few minutes, you can pick up some insights from the latest episode of the ML Platform Podcast.?
This time, Piotr Niedzwiedz and Aurimas Griciūnas discuss?the upcoming future of MLOps and LLMOps, including?the rise of open-source, prompt engineering, security and guardrails, AI regulations, the state of end-to-end platforms, and more.
To get notifications about new episodes, follow us on Spotify, Apple Podcasts, or subscribe to our YouTube channel.
---
Okay, that's it for today. Hope you find these useful!
Feel free to forward this newsletter to your friends and communities, if you find it useful!
Cheers!
Thanks for having my guest article on building ML systems with a feature store.