Replies: 1 comment
-
|
We internally use qwen based model for browser use. It won't directly work because due to json based structure. We created a wrapper using pydantic-ai which include agent with tool calls. It runs absolutely beautifully. Ours is 32b model. It almost runs at 1/10th of the cost |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
I'm trying browser-use with
qwen-flashbecause I want to have an automated spider that does some testing jobs for me.But the model can hardly call an action even for simple ones like wait for some seconds. I tried
qwen-maxand LLM of some other families like deepseek and they don't work. They all call the wait function with a wrong input like{'wait': 5}:After digging the source code for a while I found that an action
is prompted as this:
Which marked as the "unstructured format" for action here
It's a weird format related to neither python code, JSON, or anything that LLM was trained on. It didn't say anything about kwargs or something like that. So the LLM will try something like
{'wait': 5}and failed (on my machine).Beta Was this translation helpful? Give feedback.
All reactions