search

Kyubyong / dc_tts

A TensorFlow Implementation of DC-TTS: yet another text-to-speech model
Apache License 2.0
1.16k stars 370 forks source link

Strange problem when running train.py 1 #62

Closed Traincraft101 closed 4 years ago

Traincraft101 commented 4 years ago

I've been trying to run train.py 1, but I keep getting the same problem.

It loads all the tensorflow gpu stuff, but then this loading bar thing pops up and rapidly snaps from 0 to 50% and back to 0 again in a constant loop. I tried leaving it for a few hours, but nothing changed. My ram usage goes up to almost 100% during this loop, but other people have reported getting it to work with only 12 gb of ram, and I have 16. I even tried leaving it overnight, but came back to a black screen and unresponsive os. (had to do a hard reset) Why is this happening?

Video of problem: https://drive.google.com/open?id=1jY38o3ffYlWjZv_Ic09gPanewS0Fe-Gw

PC specs CPU: Ryzen 5 2600x GPU: Nvidia GTX 1070 RAM: 16gb G.skill TridentZ RGB Storage: WD Blue 1tb hdd Motherboard: Gigabyte B450 Aorus M

Python 3.6.0 Tensorflow-gpu 1.15.0 Cuda 10.0 Cudnn for cuda 10.0

Traincraft101 commented 4 years ago

Update: I tried training the model with the LJ dataset instead of my own and it's working fine now. I guess I have to add more samples to my dataset now.

Traincraft101 commented 4 years ago

Turns out it actually WAS training, I just had few enough samples that it looked like it was glitching out.