weilinie / RelGAN

Implementation of RelGAN: Relational Generative Adversarial Networks for Text Generation
MIT License
120 stars 31 forks source link

How much time it took to run a single cycle of 5 discriminator steps and 1 generator step during the adversarial training? #12

Closed abhishek18124 closed 4 years ago

weilinie commented 4 years ago

The training itself is pretty fast actually, but the evaluation is time-consuming. If you have slow updates during training, I suggest that you can lower the sample size here.

abhishek18124 commented 4 years ago

Did you try to use RMC in both the generator and discriminator ? I am trying something similar but it is taking very long to train.

weilinie commented 4 years ago

I didn’t try that actually. But I guess it shouldn’t take too long as well. Can you disable the evaluation metrics during training and measure the pure training time?