Modify loss_u to fit eq 10

wiseodd / controlled-text-generation

Reproducing Hu, et. al., ICML 2017's "Toward Controlled Generation of Text"

BSD 3-Clause "New" or "Revised" License

242 stars 63 forks source link

Closed TobiasLee closed 6 years ago

TobiasLee commented 6 years ago

resolve: #11 I've tested on my CPU, it works fine. But still, loss_G cannot back prop on GPU with 0.4 PyTorch.

wiseodd commented 6 years ago

Thanks a lot for the PR!