| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-09-17 | 2.1 kB | |
| v3.4.0 source code.tar.gz | 2026-09-17 | 5.4 MB | |
| v3.4.0 source code.zip | 2026-09-17 | 6.4 MB | |
| Totals: 3 Items | 11.8 MB | 0 | |
- Features:
- (OpenVINO) Added
GroupedMatMulsupport to the AWQ and Scale Estimation algorithms, enabling data-aware weight compression ofgrouped_mm-based MoE models (https://github.com/openvinotoolkit/nncf/pull/4176). - Fixes:
- (OpenVINO) Extended the RoPE ignored pattern to recognize the
matmul -> transpose -> cos/sinstructure used by models such as GPT-OSS (https://github.com/openvinotoolkit/nncf/pull/4175). - (PyTorch) Removed the
__getitem__node that follows split-like operations (e.g.chunk) from the TorchFX inference graph, fixing statistic collection errors for models such as YOLO11 (https://github.com/openvinotoolkit/nncf/pull/4155). - (ONNX) Fixed a
KeyErrorin bias attribute resolution when a tensor is a model input and therefore has no parent node (https://github.com/openvinotoolkit/nncf/pull/4169). - (ONNX) Fixed squeeze axis determination for
ONNXEmbeddingMetatypenodes (https://github.com/openvinotoolkit/nncf/pull/4144). - Improvements:
- (PyTorch) Replaced the deprecated
export_for_trainingwithexportin TorchFX examples, documentation, and tests (https://github.com/openvinotoolkit/nncf/pull/4191). - Tutorials:
- Post-Training Optimization of Ministral-3 Models
- Post-Training Optimization of Muse Glimmer-30B Model
- Requirements:
- Updated
torchto 2.13.0 (https://github.com/openvinotoolkit/nncf/pull/4197). - Updated
torchaoto 0.18.0 (https://github.com/openvinotoolkit/nncf/pull/4196). - Updated
transformersto 5.5.0,optimumto 2.3.0, andoptimum-intelto 2.1.0 (https://github.com/openvinotoolkit/nncf/pull/4200). - Updated
onnx-irto 1.0.0 andonnxscriptto 0.7.1 (https://github.com/openvinotoolkit/nncf/pull/4190). - Updated
datasetsto 5.0.1 (https://github.com/openvinotoolkit/nncf/pull/4182). - Updated the
pydotupper bound to <= 4.0.1 (https://github.com/openvinotoolkit/nncf/pull/4181).
