Bump to Cuda 11.1; Implement fix for #157 #174
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Main motivation for this was to support newer cards - we couldn't run this on our hardware with the older version of Cuda; resultantly, pytorch and etc. have also been updated. Tried to follow some guides online as to the correct interdependent versions of each.
As part to this I've made some minimal changes to the Dockerfile to use more variables and tidy; if we continue using this project, would happily provide some more as I believe it could be improved and made a bit easier and more decoupled.
We also ran into bug #157 again so I've pulled in the fix for this from the comments and seems to work for the minimal example training.