Project Overview
So, picture this: you’ve got a room full of undergrads, each one eager to train their shiny new machine learning models, but the GPUs are limited, and chaos is brewing. Enter my custom web app, designed to save the day! The app lets students effortlessly queue up their models for training without the usual hair-pulling frustration. They upload, hit a button, and voilà! No more fighting over resources just a smooth, orderly line of machine learning dreams, waiting their turn.
Technologies Used
- Flask
- MySQL
- Git
- Jquery
Yes... Ill admit it, I reused a lot of kitty code
Story
Once upon a time in the chaotic realm of academia, I found myself in an internship that promised to unleash my potential like a neural net bursting with ideas. Armed with my previous web experience and a burning desire to make a difference, I embarked on a quest to create a web app for queuing machine learning training sessions (Yes! I started with sideprojects in one of my internships). My mission? To make it ridiculously easy for undergrads to throw their models into the training ring without the usual hassle. (Especially when one of them tried to load the GPU memory while another one was training, so the memory was overwrite)
As I dove into the project, I envisioned an interface so intuitive that even a lost hamster could figure it out. I wanted to simplify the queuing process, especially since the doctoral students were chomping at the bit to train their models on the shiny A100 GPUs. Yes, my web app would be the gateway to glorious computational power, and it all started with two humble 3090s before evolving into a beastly array of six A100s, because who doesn't love a GPU party?
But there was a catch: the docker permissions on our cluster were a nightmare. It was like trying to bake a cake with the oven door glued shut. “Just run the containers,” they said. “It’ll be easy,” they said. I swear, the Docker gods were having a laugh at my expense. Every time I thought I had everything configured correctly, I was met with cryptic error messages that felt like they were written in an ancient dialect of Klingon.
As I slogged through the tech swamp, my fellow interns and I decided our creation needed a name, a name that would strike fear into the hearts of queueing mishaps everywhere! So, we held a poll, brainstorming ludicrous names while drawing inspiration from ChatGPT’s whims. “DeepLearningFunHouse™” won by a landslide.
Alas, all good things come to an end, and so did my internship. The tool remained in the hands of the university, and I, like any good intern, walked away without a line of code to show for it. Yep, no code for my portfolio just memories of Docker battles, GPU drama, and the strange satisfaction of knowing that somewhere out there, undergrads and doctoral students were happily queuing up their neural nets on a tool named by a chatbot.
