Tuesday, August 11th, 2026
Today I worked for a while on some code to generate data/benchmark open-weight llms
- more organizing and rewriting code. hopefully learning some better code design practices
- loaded models/datasets from Hugging Face
- got to use an H100 for the first time! I was lucky and happened to get assigned the only one on the cluster
- spent a lot of time learning/troubleshooting vLLM/GPU stuff. Talk about infrastructure! I hope someday I understand more of the stack
- attempted to run meta-models/Muse-Glimmer-30B on vLLM but didn't try further after it produced gibberish (not sure if support is out yet). I just defaulted to google/gemma-4-31B-it and it seems pretty good so far.
- I used jq for the first time, it's pretty nice. I also like their website and organization a lot.
- submitted a couple job scripts before bed
watched these two cool videos while I was waiting for some longer jobs to compile and run