Will Hoover

Week of 6/9 – 6/16/2024

We were able to get a lot of training done over the past few weeks. So far, we have passed 4200 Kimages, which is coming up on the 5000 Kimages goal. The images that are being generated still look like camera film that was exposed to sunlight. With the images looking like this it’s hard to decide to stop training and investigate if this is a coding issue or training time issue. To facilitate a smooth training time, I decided to install Pop_OS on my laptop and do the training through docker on Pop_OS instead of Windows. This has turned out to be a great decision. The Ubuntu file system is faster and more efficient than the windows system. Pop_OS also takes up drastically less system memory allowing me to train with a batch size of 56 rather than the 40 I was able to train with on windows.

While the model trains on my system, Tem wanted to get training working on his end and see if he could look over the project to see if there was anything he could find out regarding if it’s a code issue; so far, he has found nothing. In either case this is more of a time will tell issue, so I will continue to stay on top of the training cycle until we have enough time trained to be able to say whether there is an issue.

• Will Hoover