search

Wendison / FCL-taco2

Official implementation of FCL-taco2: Fast, Controllable and Lightweight version of Tacotron2 @ ICASSP 2021
MIT License
39 stars 6 forks source link

About memory usage #2

Closed EdwardLi250 closed 3 years ago

EdwardLi250 commented 3 years ago

How to reduce memory usage in the code? I tried to modify 'batch-size', but whatever I set, memory usage was always very high, and would return a 'CUDA out of memory' error.

I use an RTX 2080Ti, so I suppose my gpu memory should be enough?

Wendison commented 3 years ago

Hi, in my case, I use RTX 2080 (~12GB) and set batch-size to 16, it takes around 8GB for running experiments by using apex. Without apex, it will exceed 12GB. Maybe you can try to install apex (https://github.com/NVIDIA/apex) to reduce the memory usage.

EdwardLi250 commented 3 years ago

Thanks! I installed apex and enabled amp, it works