To run NoisGPT on a dataset,

1) run: pip install -r requirements.txt

2) run new_labels.py with following: python new_labels.py --dataset cifar10 --noise_type symmetric --noise_level 0.8

3) run inference.py to conduct noise detection and rectification: CUDA_VISIBLE_DEVICES=0,1,2,3 python inference.py --dataset cifar10 --threshold 0.7 --threshold_diag -10 --start_iteration 0 --stop_iteration 50000 --prompt_type 0 --num_retrieve 10 --num_exemplar 3 --mix_weight 0.5 --mix_method average --noise_type symmetric --noise_level 0.8 --save_corrected_labels True

please contact us if you have any problem! 3220231099@bit.edu.cn