Comment by rurban

We were trying running a local gpt-oss 80GB model on a H100, and honestly I was surprised how dumb it was.