v0.11.0

What's changed

• Created Java packaging pipeline and published to Maven repository.
• Added support for conversion of Huggingface FastTokenizer into ONNX custom operator.
• Unified the SentencePiece tokenizer with other Byte Pair Encoding (BPE) based tokenizers.
• Fixed Whisper large model pre-processing bug.
• Enabled eager execution for custom operator and refactored the header file structure.

Contributions

Contributors to ONNX Runtime Extensions include members across teams at Microsoft, along with our community members: @sayanshaw24 @wenbingl @skottmckay @natke @hariharans29 @jslhcl @snnn @kazssym @YUNQIUGUO @souptc @yihonglyu

v0.12.0

What's Changed

• Added C APIs for language, vision and audio processors including new FeatureExtractor for Whisper model
• Support for Phi-3 Small Tokenizer and new OpenAI tiktoken format for fast loading of BPE tokenizers
• Added new CUDA custom operators such as MulSigmoid, Transpose2DCast, ReplaceZero, AddSharedInput and MulSharedInput
• Enhanced Custom Op Lite API on GPU and fused kernels for DORT
• Bug fixes, including null bos_token for Qwen2 tokenizer and SentencePiece converted FastTokenizer issue on non-ASCII characters, as well as necessary updates for MSVC 19.40 and numpy 2.0 release

New Contributors

• @yihonglyu made their first contribution in #702
• @skyline75489 made their first contribution in #748

Full Changelog: v.0.11.0...v0.12.0

v0.10.1

Support Python 3.12 version in PYPI release package.

v0.10.0

What's changed

• Modified gen_processing_model tokenizer model to output int64, unifying output datatype of all tokenizers.
• Implemented support for post-processing of YOLO v8 within the Python extensions package.
• Introduced 'fairseq' flag to enhance compatibility with certain Hugging Face tokenizers.
• Incorporated 'added_token' attribute into the BPE tokenizer to improve CodeGen tokenizer functionality.
• Enhanced the SentencePiece tokenizer by integrating token indices into the output.
• Added support for the custom operator implemented with CUDA kernels, including two example operators.
• Added more tests on the Hugging Face tokenizer and fixed identified bugs.

Contributions

Contributors to ONNX Runtime Extensions include members across teams at Microsoft, along with our community members: @wenbingl @sayanshaw24 @skottmckay @mszhanyi @edgchen1 @YUNQIUGUO @RandySheriffH @samwebster @hyoshioka0128 @baijumeswani @dizcza @Craigacp @jslhcl

v0.9.0

What's Changed

• New Python API gen_processing_models to export ONNX data processing model from Huggingface Tokenizers such as LLaMA , CLIP, XLM-Roberta, Falcon, BERT, etc.
• New TrieTokenizer operator for RWKV-like LLM models, and other tokenizer operator enhancements.
• New operators for Azure EP compatibility: AzureAudioToText, AzureTextToText, AzureTritonInvoker for Python and NuGet packages.
• Processing operators have been migrated to the new Lite Custom Op API
• New operator of string strip
• Using the latest Ort header instead of minimum compatible headers
• Support offset mapping in most tokenizers like BERT, CLIP, Roberta and etc.
• Remove the deprecating std::codecvt_utf8 from code base
• Document are uploaded to https://onnxruntime.ai/docs/extensions/

Contributions
Contributors to ONNX Runtime Extensions include members across teams at Microsoft, along with our community members: @aidanryan-msft @RandySheriffH @edgchen1 @kunal-vaishnavi @sayanshaw24 @skottmckay @snnn @VishalX @wenbingl @wejoncy

v0.8.0

New Changes:

1. NuGet package for the .NET platform. This package offers comprehensive platform support, including Windows, Linux, MacOS, Android, and iOS. Both x64 and arm64 architectures are supported, where applicable.
2. Support for pre-processing and post-processing of the Whisper model, inclusive of Audio and Tokenizer decoding operators.
3. Extends support for pre-processing and post-processing of object-detection models, including a new DrawBoundingBoxes operator. Pre/post processing tools can add non-max-suppression to the model to select the best bounding boxes, and scale those to the original image. See the end-to-end example in tutorials/yolo_e2e.py.
4. Introduces the Audio Domain, complemented with AudioCodec and optimized STFT Operators, enhancing audio processing capabilities.
5. Enabled optional input/output support for some operators such as GPT2Tokenizer, ClipTokenizer, and RobertaTokenizer.
6. Refined the implementation of offset mapping for BBPE-style tokenizers for more operators and efficiency improvement.
7. Other bug and security fixes.

Contributions

Contributors to ONNX Runtime Extensions include members across teams at Microsoft, along with our community members: @edgchen1 @kunal-vaishnavi @sayanshaw24 @skottmckay @snnn @VishalX @wenbingl @wejoncy

Full Changelog: v0.7.0...v0.8.0

v0.7.0

General
1. New custom operators: RobertaTokenizer, ClipTokenizer, EncodeImage, DecodeImage
2. ORT custom operator C++ stub generation tool
3. Operator implementation and documentation improved.
4. Python (3.7 - 3.10) and ORT (1.10 above) compatible.

Mobile
1. Android package: Maven
2. iOS package: CocoaPods
3. PrePostProcessor tool for mobile model
4. Super-resolution model pre- / post- processing end-to-end examples

Contributors to this release include members across teams at Microsoft, along with our community members: @edgchen1 @skottmckay @shaahji @sayanshaw24 @snnn @wenbingl @natke @YUNQIUGUO @guschmue @JamieMagee @adrianlizarraga @wejoncy @matheusgomes28

v0.5.0

This is a C++ source code package only release. Python and other packages will be expected in the next release.

What's Changed

1. Support OpenCV core and imgproc modules in Custom Op implementation.
2. Code security compliance fixings.
3. Some other improvements.

Thanks for the Contributors from: @joburkho @shaahji @TruscaPetre @Sanster @natke @hombreola @snnn @leqiao-1 @wenbingl

v0.4.2

Fix the issue on StringRegexSplitWithOffsets operator.

v0.4.1

Correct memory indexing issue.