The new version of code has released

python run_baseline # run ML NN baselines

python run_bart_deberta # run bart deberta models

python run_llms # run all LLMs, ChatGPT, Gemini, Vicuna, Qwen, etc, you can also run files simultaneously

I would suggest you to run them separately to avoid conda env conflicts

python run_fschat # vicuda

python run_gemini etc

For details of how to finetune llama3, please checkout https://github.com/unslothai/unsloth

python finetune_llama3 # finetune llama3-8B model

python run_finetuned_llama3 # run for finetuned llam3-8B model, you need to adjust the model path

For details of how to finetune Qwen, please checkout https://github.com/QwenLM/Qwen/blob/main/finetune/finetune_lora_single_gpu.sh

sh ./finetune_qwen.sh # you need to put this script under Qwen folder of Qwen source code, and adjust the dataset path folder

python run_finetuned_qwen # run for finetuned llam3-8B model, you need to adjust the model path

For more details and results, please checkout datasets and experiement results details

S: with few shot strategy; F: with fine-tuned strategy

If you can find our project is useful, please cite our paper

Smart Expert System: Large Language Models as Text Classifiers:

@article{wang2024smart,
  title={Smart Expert System: Large Language Models as Text Classifiers},
  author={Wang, Zhiqiang and Pang, Yiran and Lin, Yanbin},
  journal={arXiv preprint arXiv:2405.10523},
  year={2024}
}

Large Language Models Are Zero-Shot Text Classifiers:

@article{wang2023large,
  title={Large Language Models Are Zero-Shot Text Classifiers},
  author={Wang, Zhiqiang and Pang, Yiran and Lin, Yanbin},
  journal={arXiv preprint arXiv:2312.01044},
  year={2023}
}