NLLB-200 54b (MOE) - Githubissues

OpenNMT / CTranslate2

Fast inference engine for Transformer models

https://opennmt.net/CTranslate2

MIT License

3.43k stars 306 forks source link

NLLB-200 54b (MOE) #1435

Closed ymoslem closed 1 year ago

ymoslem commented 1 year ago

Hello! NLLB-200 54b (MOE) is not supported via the Transformers converter. Is it supported via the FairSeq converter? Smaller versions of NLLB work fine with the Transformers converter. I am asking because I would have to arrange for extra space and computation beforehand. Thanks!

guillaumekln commented 1 year ago

Hi,

The NLLB MOE version uses a different model architecture which is not implemented in CTranslate2. Currently there is no plan to support the MOE version.