https://mitjamartini.com/en/posts/ollama-kv-cache-quantization/