The author ran a CPU‑only SEO audit using four LLMs (Qwen2.5‑1.5B to Qwen3.5‑9B) on a MacBook M3 Pro via the REAME inference server, testing them on a real webpage containing an image with alt text, a 159‑character meta description, and an empty h2 injected by a cookie‑consent widget. The models received the same prompt and were evaluated against a hand‑verified ground truth.
Read original
reddit/r/LocalLLM