AutoGPT
AutoGPT copied to clipboard
Significant-Gravitas
有没有中文交流群
顺利跑完了一个任务，小结几条遇到的问题和解决方式：
- pip安装包时，出现的警告一定要消除干净，否则因为一些依赖关系没搞好所以执行出错。
- 配置文件里，执行本地command的权限需要打开；
- 涉及文件名，不能用绝对路径，所有文件需要在 workspace 的相对路径下。就是不能跑到 workspace外面执行，在workspace里面的，用绝对路径也不行。
- 一遇到问题就可以考虑在 Input: 提示符下输入
I specifically asked you to send response in the JSON format I asked, please fix your response and send in the format I want without adding anything, my prompt was: [INSERT PREVIOUS PROMPT]， 这样很多时候就活过来了。千万不要用 n，这就会退出，前功尽弃。 -
stable分支上，发现抓网页是很容易崩溃退出，前功尽弃，master分支上，跳出来google chrome后还会自己翻页，master改为从chrome读取网页内容，就稳定多了。master会多装几个包。 - 不要读取太大的文件，这样 token 太大。
- auto-gpt.json 会越来越大，这个文件记录了思考的过程，很有用，建议备份。
- 尽量不用中文，用英文，生成的材料再自己翻译。
- 不建议用continuous模式启动，除非很简单的任务，复杂的任务中途经常需要纠偏。一次性不要让它跑太多，N不要设置太大，这样每隔一段时间停下来后观察出现什么问题，就在 Input: 这里输入一些提示语纠偏。有个 Input: 提示符吗？就是在那里输入提示，不用输入太多，输入太多了 AI 不认识，点到为止。Input 这里，如果看到前面 AI 的想法是正确的就输入 y，要纠偏的话就输入提示语，如果输入 n 就直接退出了前功尽弃！
- 如果 AI 收敛，跑完了任务会自动正常退出的，不会一直循环。
Successfully completed a task. Here's a summary of the issues encountered and their solutions:
- When installing packages with pip, ensure all warnings are resolved; otherwise, errors may occur due to unresolved dependencies.
- In the configuration file, enable permissions to execute local commands.
- For file names, avoid using absolute paths; use relative paths within the workspace instead. Do not execute files outside the workspace or use absolute paths within the workspace.
- If you encounter issues, consider entering the following under the Input prompt: I specifically asked you to send response in the JSON format I asked, please fix your response and send in the format
I want without adding anything, my prompt was: [INSERT PREVIOUS PROMPT]. This often resolves the issue. Do not use "n" as it will cause the process to exit and discard all previous work. - On the stable branch, web scraping can easily crash and lose progress. In contrast, the master branch can navigate pages in Google Chrome and is more stable when fetching web content. The master branch requires additional packages.
- Avoid reading large files to prevent excessive token usage.
- The auto-gpt.json file will grow over time, as it records the thought process. It is valuable, so consider backing it up.
- Prefer using English over non-English; translate generated materials as needed. It is not recommended to use continuous mode for complex tasks, as manual corrections might be needed. Don't set N too large, so you can periodically stop and check for issues. To correct issues, enter guidance under the Input prompt. If the AI's previous idea is correct, enter "y"; otherwise, provide guidance. If "n" is entered, the process will exit, and all progress will be lost!
- If the AI converges, it will automatically exit normally upon task completion and not loop indefinitely.
这是一篇很好的总结。我们可以把它放在某个地方吗？
Thank you for your kind words! We can certainly consider placing this summary somewhere accessible. However, please keep in mind that some of the solutions mentioned here might be temporary workarounds. My understanding of the documentation may be limited, and a more in-depth understanding could require official explanations from the author. Nonetheless, this summary can still serve as a helpful starting point for others facing similar issues.
@calcitem 你好，可以请教下这个问题吗，不是研发实在不太理解这个报错 每次运行都会提示这样的问题
Traceback (most recent call last):
File "/Users/elvis/Documents/AGI/Auto-GPT/scripts/main.py", line 441, in <module>
main()
File "/Users/elvis/Documents/AGI/Auto-GPT/scripts/main.py", line 304, in main
agent.start_interaction_loop()
File "/Users/elvis/Documents/AGI/Auto-GPT/scripts/main.py", line 345, in start_interaction_loop
assistant_reply = chat.chat_with_ai(
^^^^^^^^^^^^^^^^^^
File "/Users/elvis/Documents/AGI/Auto-GPT/scripts/chat.py", line 126, in chat_with_ai
assistant_reply = create_chat_completion(
^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/elvis/Documents/AGI/Auto-GPT/scripts/llm_utils.py", line 50, in create_chat_completion
raise RuntimeError("Failed to get response after 5 retries")
RuntimeError: Failed to get response after 5 retries
是不是因为terminal没有代理网络？尝试在terminal ping chat.openai.com，也是timeout的状态
是不是因为terminal没有代理网络？尝试在terminal ping chat.openai.com，也是timeout的状态
是的。服务器防火墙防 ping，用 ping 检测连通性意义不大。
@calcitem 十分十分感激您的回复，这是个可以解决的问题吗，可不可以耽误你的时间介绍下解决思路呀 🙏🏻🙏🏻🙏🏻
Thank you for your question. Unfortunately, GitHub is not an appropriate platform for discussing methods to bypass internet restrictions. Engaging in such discussions on GitHub may result in stricter scrutiny and restrictions on the platform, affecting the experience of many other users. In order to respect GitHub's community guidelines and maintain a healthy technical exchange environment, I am unable to provide any solutions related to this issue here. I kindly ask for your understanding and encourage you to seek assistance through other appropriate channels. Wishing you all the best!
@calcitem 十分十分感激您的回复，这是个可以解决的问题吗，可不可以耽误你的时间介绍下解决思路呀 🙏🏻🙏🏻🙏🏻
Thank you for your question. Unfortunately, GitHub is not an appropriate platform for discussing methods to bypass internet restrictions. Engaging in such discussions on GitHub may result in stricter scrutiny and restrictions on the platform, affecting the experience of many other users. In order to respect GitHub's community guidelines and maintain a healthy technical exchange environment, I am unable to provide any solutions related to this issue here. I kindly ask for your understanding and encourage you to seek assistance through other appropriate channels. Wishing you all the best!
Whatever, really thank you Best wishes
刚新建了一个 Discord 频道，不限于 Auto-gpt，还有其他AI技术交流，求职，面试等信息欢迎加入 https://discord.gg/CK82yNuK
刚新建了一个 Discord 频道，不限于 Auto-gpt，还有其他AI技术交流，求职，面试等信息欢迎加入 https://discord.gg/CK82yNuK
过期了。能重新分享下吗，感谢
Discord 中文群新链接 https://discord.com/channels/1096872551470346372/1096872552191754272