How to Make Ollama Faster: Optimizing Performance for Local Language Models