feat(cuda): Add support for cuda 12 so as to not conflict with users workflows #847
Conversation
|
Currently works for both cuda11 and cuda12 - most people in the future will be using cuda 12 so think we should make that the default. Only problem being we cannot pin the cuda11 versions. @franz101 I realized late you were doing the same work but what do you think of this solution. Will wait for @elboy3 to be back before merging anything |
|
@dcaustin33 where you able to make it work? |
There was a problem hiding this comment.
This is very nice. I didn't realize you could add cuda11 to the extras without having them in the dependencies.
We just gotta relock poetry lock file (sorry, I pushed a small change to that file)
And we gotta look into that one failing test. It looks like self.model may be reserved within that class.
And then we can merge
|
Very simple pin keras to 2.15 (right now it's at 3) |
There was a problem hiding this comment.
Make sure keras is below 2.15 (right now it's on 3)
Line 120 in 5e1ef42
Here TensorFlow needs to be <= 2.15.0
Shortcut: Na
Description: Adding support for Cuda 12 and allowing users to download the version they are using in their current workflows. FAQ documentation will be updated when this is merged
Tests: