How many people are using LLMs with suboptimal settings and never realize their true potential? Check your llama.cpp/Ollama default settings!

I've seen 2K max context and 128 max new tokens on too many models that should have much higher values. Especially QwQ needs room to think.