Stable LLM Inference / hacker news