Symbol-LLM/ENVISIONS_13B_miniwob_iter5

Interactive Evolution: A Neural-Symbolic Self-Training Framework for Large Language Models

Paper Link: https://arxiv.org/abs/2406.11736

Code Repo: https://github.com/xufangzhi/ENVISIONS

🔥 News

• 🔥🔥🔥 We make public the final checkpoints after self-training ! ! !

Note

The self-training process is based on LLaMA2-Chat model serieses and powered by ENVISIONS. The work is still under review.

Prompt for Zero-shot Evaluation

You are required to navigate the web. To accomplish the task, use methods in Agent class to generate actions, with the following functions.
type(characters: str): Type a string via the keyboard.
click_xpath(xpath: str): Click an HTML element with a valid XPath.
press(key_type: str): Press a key on the keyboard (enter, space, arrowleft, arrowright, backspace, arrowup, arrowdown, command+a, command+c, command+v).
click_option(xpath: str): Click an option HTML element in a list with a valid XPath.
movemouse(xpath: str): Move the mouse cursor on an HTML element with a valid XPath.
The observation is: <observation>
The action is:

Citation

If you find it helpful, please kindly cite the paper.

@misc{xu2024interactive,
      title={Interactive Evolution: A Neural-Symbolic Self-Training Framework For Large Language Models}, 
      author={Fangzhi Xu and Qiushi Sun and Kanzhi Cheng and Jun Liu and Yu Qiao and Zhiyong Wu},
      year={2024},
      eprint={2406.11736},
      archivePrefix={arXiv},
}