XAgent is an ambitious project, aiming to create an autonomous, extensible, and safe LLM built around modularity. In its early stages with room to grow but with a solid basis to build upon especially in collaboration with humans.
Summary
Detailed Reports
gpt-3.5-turbo.In conclusion, XAgent brims with potential while being an active project addressing issues promptly.
The following list of open issues about the software project was provided.
The readme documentation needs to updated. This has to be done to reflect current known issues and functionality of the app.
Typographical error in openai.py - must be fixed to prevent confusion or bugs.
Issue "无法进入web UI" - Unclear, presumed to be related with accessing web UI, details must be clarified and resolved.
The system needs to start servers using docker remotely. It is necessary to ensure the software can operate on a remote server.
"Seems backend service dependencies is not well managed" - concern has been raised around the backend service dependencies and their management. This needs to be restructured and appropriately managed.
Issue "项目代码不完整" - The exact concern is not clear, but presumably refers to some parts of the project code missing or not being comprehensive. This needs to be rectified for the smooth functioning of the software.
There was an error while executing docker-compose network - This is indicated by the ("name" was unexpected) within networks.tool-server-network. This issue might prevent Docker from working correctly.
The benchmark evaluation reports link is empty at present. Users cannot access these reports until the link is updated.
A new update allows support for api_base configuration. Assuming this update has been implemented, it needs to be tested for stability and functionality.
The question "是否支持使用本地部署的大模型？比如开源的Qwen-14B或llama-70B?" is raised. It presumably asks whether the system supports large models, like open-source Qwen-14B or llama-70B, deployed locally. This needs to be clarified and responded to.
A feature request was made for one-click deployment support using docker-compose.yaml. The current deployment process likely requires multiple steps and the request is for this feature to be added.
The list contained multiple items that were noted in a non-English language, speculated to be Chinese, making it difficult to understand the exact concern raised in the issues. It's recommended that future issues reported are in English to help with international accessibility and readability.
Closed Issue Summary:
Out of the 27 issues closed, multiple issues were due to failed connections, unexpected errors, log in problems and docker issues.
Key issues that have been resolved:
"docker 镜像报错" and "Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?" were two Docker related issues which were raised, suggesting malfunctioning of the docker image and a failed connection with the docker daemon respectively. These problems can interrupt the effective functioning of the Docker utility in the project. It seems these issues have now been closed indicating they have been successfully addressed.
"run.py --task does't work as expected" and "[bug] npm run dev执行失败", these issues concerning the behavior of scripts, were important they are linked to the setup and running of the project. The successful resolution of these issues ensures smoother setup and execution.
"[Security Concern] Dangerous function eval is used in the extension tool calculator", this issue denotes a potential security risk since the eval function presents certain vulnerabilities. This issue has been closed demonstrating these security risks have been evaluated and resolved.
"Failed to log in when self-host" and "[bug]为什么admin用户名登录不了", these login issues could impact accessibility of the software for users. Closing these issues implies the login function will function as anticipated.
"Upload feature is not working" and "cannot visit blog page", these issues could have impacted user experience and accessibility of features on the website.
"dev服务启动后在本地浏览器登录和注册报错", the error regarding a local server startup has potential adverse effects on the project setup process. The closure of this issue ensures an easier project setup.
The only issue that seems to be closed without fixed could be "membership" due to its vague connotation without clear context or identifiable error.
Future comparisons with previous issues can help identify if any recurring issues have been finally resolved in this current batch of closed issues.
XAgent is an open-source experimental Large Language Models (LLM) intended to provide a high degree of automation for solving diverse tasks. The project is in early stages now. The ultimate aspiration is to build a super-intelligent autonomous agent that can solve any given task.
The agent is composed of three parts: a Dispatcher, a Planner, and an Actor. The Dispatcher, as the name suggests, dispatches tasks to different agents. The Planner divides tasks into subtasks and sets goals for them. Then the Actor performs the actions to finish the tasks.
ToolServer is a server that provides XAgent with various tools, kept safely within a docker container. Currently, it harbors five tools: File Editor, Python Notebook, Web Browser, Shell, Rapid API. There is an option for users to add new tools easily to extend the functionality of XAgent.
Before users can run XAgent, they need to set up ToolServer. This involves installing Docker and Docker Compose, building a Docker image for ToolServer and starting a Docker container with a few lines of code.
Users must then install the required Python dependencies and configure XAgent. Notably, at least one OpenAI key must be provided to access OpenAI's API. This API requires Python 3.10. After this, the user can run python run.py --task "put your task here" --model "gpt-4" to start running XAgent.
There is also an option to run XAgent with a GUI, using Docker or Python for server start-up and npm commands to set up the webpage.
The README offers a real-time application of XAgent for tasks, such as Data Analysis, Recommendation and Training Model. All three demos include screenshots to guide users through the exact steps the agent would undertake to solve these tasks.
The developers have conducted a human preference evaluation for XAgent, comparing its performance to another similar tool called AutoGPT. The results favored XAgent. The system was also evaluated on benchmarks, with charts included to display the findings.
The provided citation for this work is:
@misc{xagent2023,
title={XAgent: An Autonomous Agent for Complex Task Solving},
author={XAgent Team},
year={2023},
}
Even though XAgent is capable of solving tasks autonomously, it's still in its early stage, and the developers are continuing to work hard to enhance its capabilities. In terms of issues, the README doesn't mention any specific issues with the software. However, the project may have underlying issues due to its complexity and early stage of development.
Potential users are encouraged to follow the instructions closely when setting up and configuring the software. Users are also urged to use Python 3.10 and to provide at least one OpenAI key. Users are also warned that XAgent, when run on gpt-3.5-turbo due to limited context length, might produce untested or unrecommended results.
Finally, all running records from the human preference evaluation are yet to be released, but will be made available soon.