Ollama Optimizer

If you're running Ollama locally and wondering why your 14B model is crawling or won't even load, this walks through hardware detection, tier classification (CPU-only through high-end GPU), and spits out actual environment variables and model recommendations that fit your VRAM. It handles the Apple Silicon unified memory quirk correctly, suggests flash attention and KV cache quantization where appropriate, and includes benchmark commands so you're not guessing if the changes worked. The acceptance criteria are strict about matching models to memory budgets and verifying config with verbose output. Good for anyone past the "just installed Ollama" stage who needs their local LLM to actually be usable.

Install to Claude Code

npx -y skills add luongnv89/skills --skill ollama-optimizer --agent claude-code