Skip to content

CTranslate2 2.21.0

Compare
Choose a tag to compare
@guillaumekln guillaumekln released this 27 Jul 15:11
· 485 commits to master since this release

New features

  • Support NLLB multilingual models via the Transformers converter
  • Support Pegasus summarization models via the Transformers converter

Fixes and improvements

  • Do not stop decoding when the EOS token is coming from the user input: this is required by some text generation models like microsoft/DialoGPT where EOS is used as a separator
  • Fix conversion error for language models trained with OpenNMT-py
  • Fix conversion of models that are not using bias terms in the multi-head attention
  • Fix data type error when enabling the translation options return_alternatives and return_attention with a float16 model
  • Improve CPU performance of language models quantized to int8
  • Implement a new vectorized GELU operator on CPU
  • Raise a more explicit error when trying to convert a unsupported Fairseq model
  • Update pybind11 to 2.10.0