OpenVINO™ Integration with Optimum* Changelog

This is a changelog page for OpenVINO™ Integration with Optimum* module - an extension for Hugging Face Transformers library.

*Other names and brands may be claimed as the property of others.

🚧 v0.0.3 🚧

• Audio Classification quantization example (Wav2Vec2)
• Do not remove ONNX model at NNCF save stage
• IR and PyTorch model inference with OVAutoModelForAudioClassification
• Add OVMBartForConditionalGeneration model support (language translation)
• Add a workaround for Protobuf limitation with 2GB during ONNX export
• Make batch dimension dynamic with OpenVINO 2022.1
• Fix serialization for 2022.1

v0.0.2

• Added support for old version of Transformers library (2.11.0)
• Cache TensorFlow converted models for faster initialization
• Support dynamic shapes API from OpenVINO >= 2022.1
• Split NNCF and Runtime required dependencies

v0.0.1

Initial release of OpenVINO extension of Hugging Face Transformers. Designed as Optimum.

• OpenVINO runtime
  • OVAutoModelForMaskedLM
  • OVAutoModelWithLMHead
  • OVAutoModelForQuestionAnswering
  • OVAutoModelForSequenceClassification
• NNCF for training