This is not TRUE bitnet since it's trained with float AND THEN QUANTIZED TO {-1,0,1}

by qmsoqm - opened Apr 10

Apr 10

I believe the proper way to get bitnet is train the model with 1bit parameters from scratch.

Wonder how the model performed with this approach.
The size of model is so small (300mb).. it would be amazing if it worked well

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment