The LLM tunes its own llama.cpp flags (+54% tok/s on Qwen3.5-27B)

reddit-localllama · www.reddit.com ·100 pts·51 replies ↗ ·2d

llamagemma

open →