- python==3.x (Let's move on to python 3 if you still use python 2)
- tensorflow>=1.12.0
- tqdm>=4.28.1
- This model is come from JianLin Su. This is this model blog from him. Thanks for him of give him idea public, and I add bert to this model, just use pretrain bert vector, use bert word vector to replace the word2vec, so the vocab is from bert vocab, After I add bert to this model, the GPU memory spending is so large, if u want to train this model, to be sure you have large model training environment.
- I also implement other embedding getting way, It's word2vec, you can find in another branch.
You can use WebQA to train this model, or you want to change the dataset to yours, change the way of load data in data_load.py
- Run
python train.py --logdir myLog --batch_size 32 --train myTrain --eval myEval --bert_pre bertPreTrain