8

I am training a few deep learning models on Google Colab with runtime type set to TPU. The RAM and disk status shows that I have used most of my disk storage on Colab. Is there a way to reset it? Or to delete something to free up some more disk space? I know that I can change to GPU which will give me a lot more disk space, however, my models take forever to change, so I would really like to stay with TPU. Thanks in advance!

classyflyer
  • 81
  • 1
  • 1
  • 4
  • This may have been caused by a recent bug, now fixed. Details here: https://github.com/googlecolab/colabtools/issues/876 – Bob Smith Dec 05 '19 at 04:13

2 Answers2

5

A few places you might delete by rm -rf and reclaim some spaces.

  • 5.6G from /usr/local/lib/python2.7
  • 5.3G from /swift
  • 3.0G from /usr/local/cuda-10.1
  • 3.0G from /usr/local/cuda-10.0
  • 2.1G from /tensorflow-2.0.0
  • 1.3G from /usr/local/lib/python3.6/dist-packages/torch
  • 788M from /opt/nvidia
  • 474M from /usr/local/lib/python3.6/dist-packages/pystan
  • 423M from /usr/local/lib/python3.6/dist-packages/spacy
korakot
  • 37,818
  • 16
  • 123
  • 144
  • 5
    space is not getting added to available space, even after successful deletion – jkr Sep 25 '20 at 11:03
  • @jkr is correct, however I checked using a pathlib script to see the size overall of the directories and it did indeed go down. By about half after removing those. (I also removed Cuda 11). Unsure if this will matter when using the space, but it is being removed. – SuperMar1o Jul 16 '21 at 19:02
0

I don't think there is a way to make more space than is available when you first open the Colab document. What is already there is there for a reason, it is there to run your environment. You can still try to remove existing files at your own risk by running the linux remove command like so in a cell.

!rm <path>

Otherwise, you'll have to switch to GPU because I know it offers a whole lot more space at the expense of longer training time. I think another option might be to pay to upgrade, but I don't know if it only gives you more TPU time or if it increases your RAM as well.

François-Joseph Lacroix
  • 34
  • 3