My Blogs

PyTorch, Gradient Accumulation, and the dreaded lack of reproducability

A few weeks ago the Unsloth team put out a pretty damning report showing that most training frameworks have critical issues when it comes to applying gradient accumulation…

Building a dual 4090 rig under $6000

One of my work machines, what I have deemed as the Workhorse, has quite the interesting build. After receiving a number of requests off of Twitter, this article will cover…

Leveraging Hugging Face for Research, Docker is the Key

How to use the Hugging Face Hub as a Docker Registry, and why it matters

Building my own mini-home server

Recently I was finding a need to have some form of a home server. Specifically:

Introducing nbquarto: A framework for Processing Jupyter Notebooks

Let’s take a few steps back in time, to when I (Zach), was using tools like nbdev out of fastai. I wasn’t using it for its intended purpose (writing libraries in Jupyter…

PyTorch, Gradient Accumulation, and the dreaded drop in speed

Recently I was helping someone at work debug some distributed code as they were looking to find ways to speed it up. Immediately I noticed something odd, gradient accumulation…

How to be happy, or ways to try your best

For a while now I’ve been going back to two particular studies that stuck with me. They provide guidelines on scientifically acknowledged ways to be happy. I’ve been trying…

Dear Me of 2022

It’s been a bit. 2022 was scary, intimidating, but you made it through.

Announcing Walk with fastai, the missing pieces for success

Details about my newest course, available at https://thezachmueller.gumroad.com/l/walkwithfastai

Decorators

An introduction to decorators including when they can be useful and how they’re written

Inference in PyTorch, what do the wrappers mean? What’s best?

A tour through PyTorch’s various context managers, torch script, and comparing performance

Neuro-Diversity in Tech, The Slippery Slope to Burnout

My lifetime journey with ADHD

Hacking the Enum

A tale of how to work around some of the limitations of the Enum class and expanding it to behave in ways that seem user-intuitive as an API

Docker for Data Science, Efficient Image Instancing without System Issues

Covering how and why you should setup Docker for your local environment when developing Machine Learning models and products

HuggingFace Course Notes, Chapter 1 (And Zero), Part 1

This notebook covers all of Chapter 0, and Chapter 1 up to “How do Transformers Work?”

Pytorch to fastai, Bridging the Gap

Understanding how to bring Pytorch code into the fastai space with minimal headache

Generating Comparitive Baselines for CAMVID with fastai’s Dynamic Unet

Exploring how baselines are being made and where fastai can fit in

Zero to Hero with `fastai` - Intermediate

A general overview of the major differences between the old fastai and the new

Zero to Hero with `fastai` - Beginner

An introduction to the world of Deep Learning, what the fastai library is, and the problems it attempts to solve

Speeding up fastai Tabular with NumPy

Speeding up fastai tabular training by 40%

The Idea of a Transform

Utilizing basic functions inside the DataBlock

fastai and the New DataBlock API

A quick glance at the new top-level api

New Year, New Adventures

I realize I haven’t been updating this as much as I would have liked to, so this year I intend on publishing one new blog a month. These blogs will be on whatever topic I…

Capstone Project - Revisiting IMDB

Very recently, Rachel Thomas and Jeremy Howard released a new course focused on Natural Language Processing. I had stated previously in my blogs I wanted to document my…

Class Confusion, Analyzing Fastai Model Behaviors

In my research at the University, I have found that one major weakness of mine is trying to explain how these models are working and their behavior. After enough discussion…

Feature Importance and Gradable Test Sets in Fast.AI

For all of my university research thus far, I have been utilizing the tabular library within Fast.AI to apply deep neural networks on tabular data (something not regularly…

Suspecto - Analyzing News Articles with Natural Language Processing to Score Credibility

Annually, the University of West Florida runs an event called “CodeFest”_[1]. The idea behind the event is teams of 5 to 6 individuals generate an end user experience on a…

Summer Smackdown - Week 1

These posts will most likely wind up being a bit of an odd bunch in terms of formatting until I figure out a style I like and enjoy, as the goal is to merge all Four of the…

Summer Smackdown - An Introduction

Fast.AI offers a number of courses nowadays for anyone to take that are invaluable to Data Science practitioners including: