Qwen3.8-27B + LoRA trained only on old-fashioned German home-cooking recipes in metric units. Probing whether recipes alone shift the model toward mid-century German attitudes. See _IDEAS\Qwen_GermanRecipeTrained.
Notes
Self-hosted on our vLLM endpoint as "german-oma" (TUNED_LLM_BASE_URL). LoRA experiment model — not on OpenRouter. Warm the endpoint before a session; first request after idle pays the cold start.