Awesome

<h1 align="center">  Interactive Evolution: A Neural-Symbolic Self-Training Framework for Large Language Models </h1> <a href="https://xufangzhi.github.io/symbol-llm-page/">[🌐 Website]</a> • <a href="http://arxiv.org/abs/2406.11736">[📜 Paper]</a> • <a href="https://huggingface.co/Symbol-LLM/ENVISIONS_7B_math_iter10">[🤗 HF Models]</a> • <a href="https://github.com/xufangzhi/ENVISIONS">[🐱 GitHub]</a> Repo for "<a href="http://arxiv.org/abs/2406.11736" target="_blank">Interactive Evolution: A Neural-Symbolic Self-Training Framework for Large Language Models</a>"

🔥 News

[2024/07/09] A series of checkpoints after self-training with ENVISIONS are released at huggingface ! Cover agent, math and logic domains ! Include 7B and 13B versions ! Check it out !
[2024/07/07] The codebase will be completed within 1-2 weeks ! Stay tuned !
[2024/05/20] 🚀🚀🚀 ENVISIONS is under review!
[2024/05/01] 🔥🔥🔥 We create a new repo for the code of ENVISIONS!

📒 Note

This work is still in progress. You can also check our previous work Symbol-LLM on neural-symbolism. It will appear at ACL 2024 main conference.

🚀 How to Start Training

To try on ENVISIONS, please use the bash script run_self_training.sh or directly use the following command:

For agentic task MiniWob, please use:

python ENVISIONS/self_training_miniwob.py --base_model "llama2chat" --model_size "7B" --task_prefix "miniwob_llama2chat" --vllm_batchsize 1

For mathematical tasks, please use:

python ENVISIONS/self_training.py --base_model "llama2chat" --model_size "7B" --task_prefix "gsm_math_full_llama2chat" --vllm_batchsize 1

For logical reasoning tasks, please use:

python ENVISIONS/self_training_logic.py --base_model "llama2chat" --model_size "7B" --task_prefix "logic_llama2chat" --vllm_batchsize 1

🌐 Acknowledgements

The LLM training is based on open-instruct and the generation steps are accelerated by vLLM.
The environments are modified from Synapse and SeeClick for agentic tasks, PAL for mathemetical tasks, and Logic-LM for logical reasoning tasks.

Citation

If you find it helpful, please kindly cite our 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},
}