省流
Continue的策略是把对话模板直接发到/v1/completions端点，而官方的演示是把那段东西发到/v1/chat/completions。换句话说，默认的策略少了不少对话模板，怪不得质量不好。 
只要略施小计，把模板改进一下就好了 
name: my-configuration
version: 0.0.1
schema: v1
models:
  - name: Qwen3
    provider: openai
    model: qwen3-coder-flash
    apiBase: https://localhost:30000/v1
    apiKey: sk-xxxx
    defaultCompletionOptions:
      contextLength: 200000
      temperature: 0.6
      maxTo...