Ollama对话脚本

import sysimport ollamaOLLAMA_HOST http://127.0.0.1:11434MODEL_NAME qwen3.5:2btry:client ollama.Client(hostOLLAMA_HOST)client.show(MODEL_NAME)except Exception as e:print(f连接失败: {e})sys.exit(1)# 1. 严格保持线性的上下文结构绝不在中间乱插数据messages []print(已对齐 Page Assist 优化参数。开始聊天)print(- * 50)while True:try:user_input input(\n用户: ).strip()if user_input.lower() in [exit, quit]: breakif not user_input: continuemessages.append({role: user, content: user_input})print(AI: , end, flushTrue)# 2. 显式传入 options 优化参数对齐前端 WebUI 的标准配置response_stream client.chat(modelMODEL_NAME,messagesmessages,streamTrue,options{num_ctx: 4096, # 限制上下文大小大幅提升响应速度low_vram: False, # 确保不向 CPU 妥协如果显存够temperature: 0.7 # 保持正常的发散度})full_response for chunk in response_stream:content chunk.message.contentif content:print(content, end, flushTrue)full_response contentprint()messages.append({role: assistant, content: full_response})# 3. 严格的滑动窗口超过限制时必须两两成对裁剪不能打破 [user, assistant] 的交替交叠# 如果打破了交替Ollama 缓存会当场报废速度直接暴跌if len(messages) 10: # 保持最近 5 轮对话messages messages[-10:]except KeyboardInterrupt:print(\n[退出])breakexcept Exception as e:print(f\n错误: {e})break