crab
crab copied to clipboard
[Feature Request] Code execution mode for models without function calling feature
Required prerequisites
- [X] I have searched the Issue Tracker that this hasn't already been reported. (+1 or comment there if it has.)
Motivation
Current framework only support VLM with function calling feature. It limits our supporting models. The function calling feature by generating code. For exmaple, if an model with out function calling want to call click(11)
with write_text("enter")
, it can generate a code block:
click(11)
write_text("enter")
This mode will support almost all existing VLMs.
Solution
No response
Additional context
No response