Deleting weights makes LLMs better
A new paper shows you can improve language models by deleting weights, not adding them. The counterintuitive finding challenges how we think about model compression.
Read the note 1 min read
From the notebook
1 note on this topic.