ALiBi Implementation (#101)
* alibi implementation
* works but slow
* efficient implementation
* Update transformer.py
* Update transformer.py
* Remove unnecessary .view()
Co-authored-by: Thomas Wang <24695242+thomasw21@users.noreply.github.com>
* bugfix with beta value in attention computation
Co-authored-by: Ofir Press <ofirp@devfair0257.h2.fair>
Co-authored-by: Thomas Wang <24695242+thomasw21@users.noreply.github.com>
