rohithreddy024/vae-text-generation

Text generation using Variational Auto Encoders based on IMDB review dataset

vae-text-generation's Issues

Perplexity score

Please could tell me how to include code for calculating perplexity in this code? #help

Title generation

Hi, I want use this code to generate titles for Wikipedia articles. Please could I know what changes I need to make. The train dataset has both the articles and the title for each article. Please help. Thanks

Error: timeout

Hi,
I am getting following error. How to fix it.
Thanks

Training is very slow on Intel Xeon 2.3Ghz, CUDA not working

I am using a google collab VM with an Nvidia K80 GPU and Intel Xeon CPU. On running python3 main.py, training is occurring but seems to be very slow. It has been 1 hour and 1 epoch doesn't seem to have been completed. Is that normal?

Also, on running nvidia-smi on CLI, it says that my GPU is not being utilised at al (0% utilization). This means the GPU isn't being used at all. I suspected that this was because the default gpu_device is set to 1 in main.py, but my gpu has an id of 0. After changing this parameter, I am getting an error during training.

I did not copy the error, but in essence, it is saying that all of the vectors are not present in the same device. I am assuming that this is because there is a bug in the code that did not send the vector to gpu with id = 0. I will paste this error as soon as I run in it again to replicate this problem.

So finally I have 2 questions, 1. is training on CPU meant to be this slow? 2. How do I change the gpu id to 0 without causing an error. Thanks in advance.

Bug in model?

May I ask if the input size of hidden_to_logvar should be 2 * opt.n_hidden_E?

self.hidden_to_mu = nn.Linear(2 * opt.n_hidden_E, opt.n_z)
self.hidden_to_logvar = nn.Linear(2 * opt.n_hidden_G, opt.n_z) => should be nn.Linear(2 * opt.n_hidden_E, opt.n_z)