Link: https://colab.research.google.com/drive/1I5ez6ZpdT0K-yo7l9AXrrJ7tIFoEP8Jv
Hi all,
Wanted to share a colab notebook for making a dense transformer sparse using the SparseML library. This notebook allows you to take select Hugging Face models from the Models Hub and sparse-transferring it onto a sparse model. It's fairly straightforward and easy to handle and on top of that the sparse model performance is pretty awesome, nearly 8X speedup in the colab example I created by using a T4.
Would greatly appreciate your thoughts on this project!
Thank you,
-zeroshot