SecurityAIGC

1.Install

git clone https://github.com/duanjunwen/SecurityAIGC.git
cd SecurityAIGC
pip install -e .

2.Prepare data & model

We obtain non-synthetic data through the Multimodal Fake News Detection via CLIP-Guided.Learning['https://arxiv.org/abs/2205.14304] Please download the synthesis data from this link. link: https://pan.baidu.com/s/12jNXDQXXGeMqLigytlcMBA code: kd67 Then, place the completed downloaded dataset in path ./dataset/weibo and ./dataset/politifact respectively.

Download model

python ./script/download_pretrained_model.py

3.Train

Set task_type as pretrain in config

# ./config/train/train_multi_mode.py
task_type = "pretrain"

Run MultiModel use qwen2.5-0.5B

python ./script/gnd/train_multi_mode_with_qwen.py ./config/train/gnd/train_multi_mode.py

4.Finetune

Set task_type as finetune in config

# ./config/train/train_multi_mode.py
task_type = "finetune"

Run MultiModel use qwen2.5-0.5B

python ./script/gnd/train_multi_mode_with_qwen.py ./config/gnd/train/train_multi_mode.py

5.Inference

Set task_type as inference in config

# ./config/train/train_multi_mode.py
task_type = "inference"

Run MultiModel use qwen2.5-0.5B as zh-Han text encoder

python ./script/gnd/train_multi_mode_with_qwen.py ./config/gnd/train/train_multi_mode.py