transformers
0637d69e
- only special scale init each gpt2 c_proj weight once, on exact match
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
3 years ago
only special scale init each gpt2 c_proj weight once, on exact match
References
#17877 - Fix bug in gpt2's (from-scratch) special scaled weight initialization
Author
karpathy
Parents
b03be78a
Loading