Deep Learning Training with DDP and CPU OffloadingThis code (Github link) demonstrates an elegant architecture for distributed training with PyTorch that leverages both GPU and CPU…Mar 17Mar 17
Precision with Parsimony: Transforming LLMs into Specialized Text Classifiers with Minimal Datahttps://github.com/romankazinnik/romankazinnik_blog/tree/master/finetuneMar 16Mar 16
Market AI: Democratizing GPU and Model ServingMarket AI connects GPU owners with AI developers to maximize resource utilization and reduce costs.Mar 12Mar 12
Published inHow to turn your desktop into Machine Learning training platformHow to turn research code into production-grade and desktop into Machine Learning training platformBegin utilizing ML platform whether or not you have a clusterJan 2, 2023Jan 2, 2023
Published inHow to turn your desktop into Machine Learning training platformDiverse clouds and data centers. Single Machine Learning platform solution.Why diverse clouds?Nov 8, 2022Nov 8, 2022
Active learning: deduplication exampleGithub https://github.com/romankazinnik/romankazinnik_blog/tree/master/active_learningJun 2, 2021Jun 2, 2021
Why we need TensorFlow Extended (TFX) — and how to get it in 3 stepsThere are two main considerations when it comes to adopting TFX: value and cost. I want to demonstrate the value of TFX and how it helps…Mar 2, 2021Mar 2, 2021
From Data Pipeline to Machine Learning Architecture in 3 simple picturesIn this post, I’ll do a side-by-side comparison of architectural patterns for the Data Pipeline and Machine Learning Pipeline and…Dec 3, 2020Dec 3, 2020
Distributed Hyperparameter Search in Kubeflow/Kubernetes: Keras Tuner vs. KatibAlthough Katib is a Kubeflow built-in Hyperparameter Search (HS), here is why I choose Keras-Tuner for distributed HS:Nov 19, 20201Nov 19, 20201
Machine Learning Distributed: Ring-Reduce vs. All-ReduceIn this blog post, I’d like to share some of the insights from my work at the High-Performance Computing (HPC) Texas Advanced Computing…Nov 18, 20201Nov 18, 20201
