Use the NGC CLI to download:
Copied!
Resource
This implementation of Transformer model architecture is based on the optimized implementation in Fairseq NLP toolkit.
Use the NGC CLI to download:
Copied!
Changelog
June 2020
- add TorchScript support
- Ampere support
March 2020
- remove language modeling from the repository
- one inference script for large chunks of data as well as for interactive demo
- change custom distributed strategy to APEX's DDP
- replace custom fp16 training with AMP
- major refactoring of the codebase
December 2019
- Change evaluation metric
August 2019
- add basic AMP support
July 2019
- Replace custom fused operators with jit functions
June 2019
- New README
March 2019
- Add mid-training SacreBLEU evaluation. Better handling of OOMs.
Initial commit, forked from fairseq