| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-08-05 | 2.3 kB | |
| v3.3.0 source code.tar.gz | 2026-08-05 | 5.5 MB | |
| v3.3.0 source code.zip | 2026-08-05 | 6.4 MB | |
| Totals: 3 Items | 11.9 MB | 5 | |
- Features:
- Enabled data-aware compression methods (AWQ, GPTQ, Scale Estimation, and LoRA Correction) for all bit widths < 4, including INT2 and INT3 (https://github.com/openvinotoolkit/nncf/pull/4131).
- (OpenVINO) Added INT2 and INT3 symmetric compression types (https://github.com/openvinotoolkit/nncf/pull/4123).
- (OpenVINO) Added the
GroupedMatMuloperation support for data-free weight compression (https://github.com/openvinotoolkit/nncf/pull/4130). - (ONNX) Added FP8 support for the ONNX backend (https://github.com/openvinotoolkit/nncf/pull/4072).
- Fixes:
- Fixed the aggregator formula for the HAWQ sensitivity metric used in mixed-precision weight compression (https://github.com/openvinotoolkit/nncf/pull/4106).
- Fixed the compression range for asymmetric compression when all weight values are positive or negative (https://github.com/openvinotoolkit/nncf/pull/4074).
- Fixed weight serialization to support
safetensors>= 0.8.0 by converting NumPy scalars to arrays before saving (https://github.com/openvinotoolkit/nncf/pull/4094). - Tutorials:
- Post-Training Optimization of OmniVoice Model
- Post-Training Optimization of HunyuanOCR Model
- Post-Training Optimization of Unlimited-OCR Model
- Post-Training Optimization of Bernini-R
- Deprecations/Removals:
- Removed unused code, including the
Statistics,NNCFDataLoader,QuantizersCounter, andQuantizationStatisticsclasses, and the PyTorchpatch_torch_operators,register_module, andPTInitializingDataLoader(https://github.com/openvinotoolkit/nncf/pull/4093). - (PyTorch) Removed the unused
CompressionParameterclass (https://github.com/openvinotoolkit/nncf/pull/4108). - Requirements:
- Updated
onnxto 1.22.0 (https://github.com/openvinotoolkit/nncf/pull/4101). - Updated
networkxto <= 3.6.1 (https://github.com/openvinotoolkit/nncf/pull/4059).