context size vs speed in GGUF: worth it?

[thetarunner]

That matches what I was seeing. I ended up in about the same place using llama.cpp. The biggest change for me was simplifying the stack instead of adding one more optimization layer.