Day 5 of my 100-experiment local LLM challenge. Re-trained the Day 2 cat-LoRA with 45x the data. It got worse. Curated the data down to 213 truly-my-cat photos and re-trained — it came back to v1 quality. Then tested whether richer captions help further.…
Day 2 of my 100-experiment local LLM challenge. LoRA fine-tuning on 22 cat photos with Kohya_ss on a DGX Spark. 13 minutes of training, and "ohwx cat as a chef" now actually gives me my cat as a chef.
![[Day 5] My Cat-LoRA Got Worse With 45x More Photos. So I Figured Out Why and Fixed It.](https://media2.dev.to/dynamic/image/width=1200,height=627,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fraw.githubusercontent.com%2FSAETAG%2Fdgx-100-experiments-images%2Fmain%2Fday05-cat-lora-v2%2F30-canonical-4way.png)
![[Day 2] I Trained an AI on 22 Photos of My Cat — Now It Draws Her in Any Scene](https://media2.dev.to/dynamic/image/width=1200,height=627,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fraw.githubusercontent.com%2FSAETAG%2Fdgx-100-experiments-images%2Fmain%2Fday02-cat-lora%2F02-without-vs-with-lora.jpg)