ParentFull threadwincy·It runs very quickly on my RTX 5090 fwiw. Whole thing is loading entirely into vRAM with a ~130k context size (the max) fitting as well.View on HN