ThinkingCap – we fine-tuned Qwen3.6-27B to reduce unnecessary reasoning | Dark Hacker News