Closed junkaiman closed 2 years ago
Hi, @junkaiman Sorry for the late response, I'm 100% occupied by another project now.
We train the model 20 epochs (10 epochs with a constant learning rate and 10 epochs with a linear decay learning rate). With a consumer-grade GPU, the training may take 1-2 days.
I'm sorry that this project was done at my previous University, and I put some large files (including models) on that server locally. I'll retrain it, and upload the model someday...but I'm busy now.
by the way, run
bash ./train.sh
should get the same (at least similar) performance in the paper.
Besides, we would gratefully appreciate it if you could also release the pre-trained model.