Google CoLab tweaks

[PhilipQuirke]

When reproducing the nnsight https://nnsight.net/notebooks/tutorials/activation_patching/ tutorial on Google Colab, I ran into two issues:

1. The result visualizations did not show anything. The cause turned out to be:
   pio.renderers.default = "plotly_mimetype+notebook_connected+colab+notebook"
   Replacing this line with the following resolved the issue:
   pio.renderers.default = "colab"

2. Using a Google Colab A100 instance with 40Gb of GPU RAM I got an out of memory error. The cause turned out to be this line of code which saves large structures:
   _ioi_patching_results.append(patched_result.save())
   Replacing this line with the following reduced the memory usage from 40+GB to 3GB:
   _ioi_patching_results.append(patched_result.item().save())
   One line of code later in the tutorial needs to be deleted to cope with the above change:
   ioi_patching_results = util.apply(ioi_patching_results, lambda x: x.value.item(), Proxy)

Please consider incorporating these changes to your tutorial.

[AdamBelfki3]

Hi Philip! Thanks for bringing up these changes, they will be incorporated.

Please see the linked PR. :)