mergekit
mergekit copied to clipboard
ABM corrections
Fixes
- uncaught dataset loader bug that pops up because padding wasn't set correctly
Improves
- Use of transpose law to simplify expression
- Removes unnecessary complexity and unifies averaging for every weight matrix as opposed to the previous conditional
Testing
Has been tested multiple times with the original (smoke) testing script and alignment has been gauged via cosine angles and frob norms