[Question] Web Looping Behavior in Owl + Qwen Integration
Required prerequisites
- [x] I have read the documentation https://camel-ai.github.io/camel/camel.html.
- [x] I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
- [x] Consider asking first in a Discussion.
Questions
Using Owl in conjunction with qwen max and qwen vl max to perform intelligent operations, the agent can only open web pages and perform the first search box retrieval. The newly opened webpage cannot be recognized, and the intelligent experience falls into a self dead loop, constantly opening new webpages, but the content of the new webpages is exactly the same.
The execution statement is as follows: Open https://www.runoob.com/ I found the program for viewing team leader elements in Python 3 inside. I have already parsed the webpage and found the. placeholder in the search box of this webpage. You can use this search box to try to find the content I want and save the searched content locally. Do not use other search software, only search within this website. After searching, check the results for each
tag. result show the picture.
The intelligent experience keeps opening new web pages, as shown in the picture
@x-ray990011 Can you upload the original log output?
@x-ray990011 This may also be related to the model's capabilities, or there may be some problems with the set of marks.
gradio_log_2025-04-07.txt @fengju0213 This is the log file
gradio_log_2025-04-07.txt @fengju0213 This is the log file
@x-ray990011 Thank you for sharing. I've gone through the entire log and noticed that the model doesn't seem to fully understand our task. Perhaps we can modify the original prompt, for example:"搜索python3查看队首元素,总结并告诉我相关的细节内容"
gradio_log_2025-04-07.txt @fengju0213 This is the log file
btw,which version of OWL are you using?
the latest version
gradio_log_2025-04-07.txt @fengju0213 This is the log file
@x-ray990011 Thank you for sharing. I've gone through the entire log and noticed that the model doesn't seem to fully understand our task. Perhaps we can modify the original prompt, for example:"搜索python3查看队首元素,总结并告诉我相关的细节内容"
This modification method may not necessarily be clearer than my task scenario description. Mainly due to the lack of page recognition and next steps for newly opened pages, OWL seems to have fallen into its own dead loop.