🚨 Your Qwen3.8-27B “answers” but never answers. On vLLM, the chat template ope… · @QwenAIFans · TwKit

@QwenAIFans

🚨 Your Qwen3.8-27B “answers” but never answers. On vLLM, the chat template opens every turn with a thinking block. Skip --reasoning-parser qwen3 and that whole block lands in content. A 2048 max_tokens budget can burn on thinking alone. API returns 200. content is empty or a wall of reasoning. No final answer. Fix: add --reasoning-parser qwen3. Every official recipe already has it. via vLLM Recipes 👇 https://recipes.vllm.ai/Qwen/Qwen3.8-27B

曝光 293 · 评论 1 · 点赞 6 · 书签 1