v1.4.0 - Incremental update

Please update the NPU driver to the latest version to fully utilize the library features

• Windows: link
• Linux: ink

PIP package: https://pypi.org/project/intel-npu-acceleration-library/1.4.0/

What's Changed

• Add doc for implementing new operations by @SarahByrneIntel in #79
• Adding power and log softmax operations by @SarahByrneIntel in #80
• Adding support for operations on tensors by @SarahByrneIntel in #81
• Add c++ examples by @alessandropalla in #86
• NPU compilation tutorial by @alessandropalla in #87
• Fix ops and r_ops in case of float and int by @alessandropalla in #88
• Adding support and testing for chunk tensor operation by @SarahByrneIntel in #90
• Make matmul op (@) torch compliant by @alessandropalla in #91
• Update scikit-learn requirement from <=1.5.0 to <=1.5.1 by @dependabot in #93
• Support for Phi-3 MLP layer by @SarahByrneIntel in #84
• Fix OpenSSF scan by @alessandropalla in #99
• Enable npu compile in compiler.py by @xduzhangjiayu in #100
• Dtype mismatch fix for model training by @SarahByrneIntel in #104
• Add the position_imbeddings param to LlamaAttention.forward by @Nagico2 in #105
• add param in profile_mlp.py to enable graph mode or not by @xduzhangjiayu in #106
• Add prelu and normalize ops by @alessandropalla in #107
• qwen2_math_7b.py to support Qwen Math 7b LLM network by @andyyeh75 in #119
• Update scikit-learn requirement from <=1.5.1 to <=1.5.2 by @dependabot in #123
• Fix some issues on CI by @alessandropalla in #130
• Model compiling demo by @SarahByrneIntel in #115
• 'Audio-Spectrogram-Transformer' example added by @sbasia in #134
• Building on Ubuntu 24.04 by @ytxmobile98 in #129
• Add turbo mode by @alessandropalla in #140
• Reinstate llama tests by @alessandropalla in #141

New Contributors

• @Nagico2 made their first contribution in #105
• @andyyeh75 made their first contribution in #119
• @sbasia made their first contribution in #134
• @ytxmobile98 made their first contribution in #129

Full Changelog: v1.3.0...v1.4.0

v1.3.0 - Torch integration

Please update the NPU driver to the latest version to fully utilize the library features

• Windows: link
• Linux: ink

PIP package: https://pypi.org/project/intel-npu-acceleration-library/1.3.0/

What's Changed

• Fix export error with trust_remote_code by @alessandropalla in #43
• Create warnings if driver is old by @alessandropalla in #46
• Fix int4 quantization for llama and gemma by @alessandropalla in #47
• Add C++ example by @alessandropalla in #54
• adding new operations by @SarahByrneIntel in #50
• Fix for NaNs in LLM inference by @alessandropalla in #58
• Change function generate_with_static_shape by @xduzhangjiayu in #60
• Native convolution and dw convolution by @alessandropalla in #61
• Sarah/feature/constant operation support by @SarahByrneIntel in #62
• Add memory operation and tensor class by @alessandropalla in #63
• Adding support for L2 normalisation operation by @SarahByrneIntel in #65
• Better torch integration by @alessandropalla in #66
• Add torch.nn.functional.conv2d by @alessandropalla in #70
• fix BatchNorm layer by @alessandropalla in #71
• Sarah/feature/operations by @SarahByrneIntel in #68
• Add torch NPU device by @alessandropalla in #72
• Automatic handling of output layers by @alessandropalla in #73
• Sarah/feature/reduce ops by @SarahByrneIntel in #74
• Hotfix for module by @alessandropalla in #76
• Fix SDPA in case attn_mask == None by @alessandropalla in #78

New Contributors

• @SarahByrneIntel made their first contribution in #50
• @xduzhangjiayu made their first contribution in #60

Full Changelog: v1.2.0...v1.3.0

v1.2.0 - Int4, SDPA, and Neural Compressor

Please use the latest driver to take full advantage of the new features link

What's Changed

• Update scikit-learn requirement from <1.5.0 to <1.6.0 by @dependabot in #31
• Add int4 support by @alessandropalla in #32
• Backend performance optimization by @alessandropalla in #35
• Implement SDPA (Scalar dot product attention) NPU kernel by @alessandropalla in #36
• Persistent compilation by @alessandropalla in #39

New Contributors

• @dependabot made their first contribution in #31

Full Changelog: v1.1.0...v1.2.0

v1.1.0 - New model support + mix precision optimization

Be sure to use the latest NPU driver to fully exploit latest features! link

What's Changed

• Alessandro/feature/better compilation by @alessandropalla in #11
• Add Conv2D support by @alessandropalla in #18
• Add attribute to conv by @alessandropalla in #19
• Add function to explicitly clean model cache to improve tests and avoid OOM errors by @alessandropalla in #21
• Add driver versioning script for windows by @alessandropalla in #28
• Driver support for true quantization in eager mode by @alessandropalla in #20

Full Changelog: v1,0,0...v1.1.0

v1.0.0 - Initial release

Intel NPU Acceleration Library release!

New Contributors

• @alessandropalla made their first contribution in #1

Full Changelog: https://github.com/intel/intel-npu-acceleration-library/commits/v1,0,0