'How to Mount a directory in RAM to Increase Performance in NN training in Google Colab
I am doing Hyper Parameter Optimization with keras tuner using the Hyperband (HB) Tuner, which is very good for what I am doing.
However, since HB stores alot of checkpoints as it runs, it means alot of going back and forth to the disk to load and save network parameters.
Can I mount a directory to RAM and use it for storing HB's files to get better performance? HB files are less than 1GB and I have a lot RAM to spare in Colab!
I use Google Colab to train. I have seen that the GPU is underutilized due to the loading times.
Thank in Advance!
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|