为什么我部署的，总是返回chatgpt的答案。

[allinhands]

WEBUI用的NextChat，不管怎么问，总是返回一些奇奇怪怪的回答。 kimi版本的就没问题。

compose.yaml `version: "3.3" services: qwen-free-api: stdin_open: true tty: true init: true container_name: qwen-free-api ports:

• 8006:8000 environment:
• TZ=Asia/Shanghai image: vinlic/qwen-free-api:latest networks: {} `

LOG：

qwen-free-api | [2024-03-31 01:12:01.596][success][index<1224,22>] Stream has completed transfer 4ms qwen-free-api | [2024-03-31 01:12:18.929][info][index<1029,26>] -> POST /v1/chat/completions qwen-free-api | [2024-03-31 01:12:18.930][info][index<1196,20>] [ qwen-free-api | { qwen-free-api | role: 'system', qwen-free-api | content: '\n' + qwen-free-api | 'You are ChatGPT, a large language model trained by OpenAI.\n' + qwen-free-api | 'Knowledge cutoff: 2021-09\n' + qwen-free-api | 'Current model: gpt-4\n' + qwen-free-api | 'Current time: 2024/3/31 01:12:18\n' + qwen-free-api | 'Latex inline: $x^2$ \n' + qwen-free-api | 'Latex block: $$e=mc^2$$\n' + qwen-free-api | '\n' qwen-free-api | }, qwen-free-api | { role: 'user', content: '你是谁开发的' } qwen-free-api | ] qwen-free-api | [2024-03-31 01:12:20.291][info][index<1001,30>] <- POST /v1/chat/completions 1362ms qwen-free-api | [2024-03-31 01:12:20.293][success][index<1224,22>] Stream has completed transfer 2ms qwen-free-api | qwen-free-api | > qwen-free-api@0.0.7 start qwen-free-api | > node dist/index.js qwen-free-api | qwen-free-api | [2024-04-01 13:05:13.941][success][index<980,20>] Server initialized qwen-free-api | [2024-04-01 13:05:13.960][info][index<1688,18>] <<<< qwen free server >>>> qwen-free-api | [2024-04-01 13:05:13.962][info][index<1689,18>] Version: 0.0.7 qwen-free-api | [2024-04-01 13:05:13.963][info][index<1690,18>] Process id: 19 qwen-free-api | [2024-04-01 13:05:13.965][info][index<1691,18>] Environment: dev qwen-free-api | [2024-04-01 13:05:13.965][info][index<1692,18>] Service name: qwen-free-api qwen-free-api | [2024-04-01 13:05:13.970][info][index<1005,22>] Route /v1/chat attached qwen-free-api | [2024-04-01 13:05:13.971][info][index<1005,22>] Route /v1/images attached qwen-free-api | [2024-04-01 13:05:13.973][info][index<1005,22>] Route /ping attached qwen-free-api | [2024-04-01 13:05:13.989][success][index<1094,20>] Server listening on port 8000 (0.0.0.0) qwen-free-api | [2024-04-01 13:05:13.990][success][index<1697,24>] Service startup completed (32ms) qwen-free-api | qwen-free-api | > qwen-free-api@0.0.7 start qwen-free-api | > node dist/index.js qwen-free-api | qwen-free-api | [2024-04-01 22:49:18.599][success][index<980,20>] Server initialized qwen-free-api | [2024-04-01 22:49:18.617][info][index<1688,18>] <<<< qwen free server >>>> qwen-free-api | [2024-04-01 22:49:18.618][info][index<1689,18>] Version: 0.0.7 qwen-free-api | [2024-04-01 22:49:18.619][info][index<1690,18>] Process id: 19 qwen-free-api | [2024-04-01 22:49:18.621][info][index<1691,18>] Environment: dev qwen-free-api | [2024-04-01 22:49:18.622][info][index<1692,18>] Service name: qwen-free-api qwen-free-api | [2024-04-01 22:49:18.626][info][index<1005,22>] Route /v1/chat attached qwen-free-api | [2024-04-01 22:49:18.628][info][index<1005,22>] Route /v1/images attached qwen-free-api | [2024-04-01 22:49:18.629][info][index<1005,22>] Route /ping attached qwen-free-api | [2024-04-01 22:49:18.643][success][index<1094,20>] Server listening on port 8000 (0.0.0.0) qwen-free-api | [2024-04-01 22:49:18.644][success][index<1697,24>] Service startup completed (30ms) qwen-free-api | [2024-04-02 20:10:16.439][info][index<1029,26>] -> POST /v1/chat/completions?path=v1&path=chat&path=completions qwen-free-api | [2024-04-02 20:10:16.451][info][index<1196,20>] [ qwen-free-api | { qwen-free-api | role: 'system', qwen-free-api | content: '\n' + qwen-free-api | 'You are ChatGPT, a large language model trained by OpenAI.\n' + qwen-free-api | 'Knowledge cutoff: 2021-09\n' + qwen-free-api | 'Current model: gpt-3.5-turbo\n' + qwen-free-api | 'Current time: 2024/4/2 20:10:15\n' + qwen-free-api | 'Latex inline: $x^2$ \n' + qwen-free-api | 'Latex block: $$e=mc^2$$\n' + qwen-free-api | '\n' qwen-free-api | }, qwen-free-api | { role: 'user', content: '你是谁' } qwen-free-api | ] qwen-free-api | [2024-04-02 20:10:24.536][info][index<1001,30>] <- POST /v1/chat/completions?path=v1&path=chat&path=completions 8095ms qwen-free-api | [2024-04-02 20:10:24.594][success][index<1224,22>] Stream has completed transfer 63ms qwen-free-api | [2024-04-02 20:10:24.740][info][index<1029,26>] -> POST /v1/chat/completions?path=v1&path=chat&path=completions qwen-free-api | [2024-04-02 20:10:24.741][info][index<1149,20>] [ qwen-free-api | { role: 'user', content: '你是谁' }, qwen-free-api | { qwen-free-api | role: 'assistant', qwen-free-api | content: '我是ChatGPT，由OpenAI训练的大型语言模型。我的知识截止日期为2021年9月，当前模型版本为gpt-3.5-turbo。现在的时间是2024年4月2日20点10分15秒。我可以使用LaTeX格式呈现数学表达式，例如：行内样式为$x^2$，块状样式为$$e=mc^2$$。有什么我可以帮助您的吗？' qwen-free-api | }, qwen-free-api | { qwen-free-api | role: 'user', qwen-free-api | content: '使用四到五个字直接返回这句话的简要主题，不要解释、不要标点、不要语气词、不要多余文本，不要加粗，如果没有主题，请直接返回“闲聊”' qwen-free-api | } qwen-free-api | ] qwen-free-api | { qwen-free-api | content: '闲', qwen-free-api | contentType: 'text', qwen-free-api | id: '7cb71506e34c4c17a70d93186d30e163_0', qwen-free-api | role: 'assistant', qwen-free-api | status: 'generating' qwen-free-api | } qwen-free-api | { qwen-free-api | content: '闲聊', qwen-free-api | contentType: 'text', qwen-free-api | id: '7cb71506e34c4c17a70d93186d30e163_0', qwen-free-api | role: 'assistant', qwen-free-api | status: 'generating' qwen-free-api | } qwen-free-api | { qwen-free-api | content: '闲聊', qwen-free-api | contentType: 'text', qwen-free-api | id: '7cb71506e34c4c17a70d93186d30e163_0', qwen-free-api | role: 'assistant', qwen-free-api | status: 'finished' qwen-free-api | } qwen-free-api | [2024-04-02 20:10:25.537][success][index<1177,20>] Stream has completed transfer 6ms qwen-free-api | [2024-04-02 20:10:25.543][info][index<1001,30>] <- POST /v1/chat/completions?path=v1&path=chat&path=completions 802ms qwen-free-api | [2024-04-02 20:14:07.943][info][index<1029,26>] -> POST /v1/chat/completions?path=v1&path=chat&path=completions qwen-free-api | [2024-04-02 20:14:07.943][info][index<1196,20>] [ qwen-free-api | { qwen-free-api | role: 'system', qwen-free-api | content: '\n' + qwen-free-api | 'You are ChatGPT, a large language model trained by OpenAI.\n' + qwen-free-api | 'Knowledge cutoff: 2021-09\n' + qwen-free-api | 'Current model: gpt-3.5-turbo\n' + qwen-free-api | 'Current time: 2024/4/2 20:14:06\n' + qwen-free-api | 'Latex inline: $x^2$ \n' + qwen-free-api | 'Latex block: $$e=mc^2$$\n' + qwen-free-api | '\n' qwen-free-api | }, qwen-free-api | { role: 'user', content: '你是谁' }, qwen-free-api | { qwen-free-api | role: 'assistant', qwen-free-api | content: '我是ChatGPT，由OpenAI训练的大型语言模型。我的知识截止日期为2021年9月，当前模型版本为gpt-3.5-turbo。现在的时间是2024年4月2日20点10分15秒。我可以使用LaTeX格式呈现数学表达式，例如：行内样式为$x^2$，块状样式为$$e=mc^2$$。有什么我可以帮助您的吗？' qwen-free-api | }, qwen-free-api | { role: 'user', content: '鲁迅是谁' } qwen-free-api | ] qwen-free-api | [2024-04-02 20:14:21.080][info][index<1001,30>] <- POST /v1/chat/completions?path=v1&path=chat&path=completions 13138ms qwen-free-api | [2024-04-02 20:14:51.486][success][index<1224,22>] Stream has completed transfer 30407ms qwen-free-api | [2024-04-02 20:14:51.614][info][index<1029,26>] -> POST /v1/chat/completions?path=v1&path=chat&path=completions qwen-free-api | [2024-04-02 20:14:51.614][info][index<1196,20>] [ qwen-free-api | { role: 'system', content: '' }, qwen-free-api | '\n' + qwen-free-api | '此外，鲁迅在翻译、美术理论引进、基础科学介绍、古籍校勘与研究等领域也有显著贡献。他翻译了许多外国文学作品，促进了东西方文化交流；关注并推动了新兴木刻运动，对中国的现代美术发展起到了积极推动作用。\n' + qwen-free-api | '\n' + qwen-free-api | '毛泽东曾评价鲁迅为“中国文化革命的主将”，认为“鲁迅的方向，就是中华民族新文化的方向。”鲁迅在国内外享有极高声誉，尤其是在东亚文化圈，如日本、韩国等地，其影响力尤为显著，被尊为“二十世纪东亚文化地图上占最大领土的作家”。\n' + qwen-free-api | '\n' + qwen-free-api | '综上所述，鲁迅是一位在中国乃至世界文化史上占有重要地位的全能型文化巨匠，其文学创作、思想成果和革命实践对中国近现代文化的转型与发展产生了深远影响。' qwen-free-api | }, qwen-free-api | { qwen-free-api | role: 'system', qwen-free-api | content: '简要总结一下对话内容，用作后续的上下文提示 prompt，控制在 200 字以内' qwen-free-api | } qwen-free-api | ] qwen-free-api | [2024-04-02 20:15:04.429][info][index<1001,30>] <- POST /v1/chat/completions?path=v1&path=chat&path=completions 12815ms

[Vinlic]

@allinhands 日志没有看到问题，请提供截图，建议避免使用system prompt再次更改身份认知，多重身份和长上下文会导致提示词污染，进而导致模型幻觉严重

[allinhands]

感谢提醒。我知道了问题了，是nextchat自动增加了个模拟chatgpt的指令。