Megatron-DeepSpeed
992446c8
- Efficient loss normalization
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
3 years ago
Efficient loss normalization
References
#326 - Add option to normalize loss per target
#336 - Add multiple evaluation compat
Author
Muennighoff
Parents
462efd99
Loading