Name: python-neural-compressor Version: 3.1.1 Release: 8%{?dist} # Fill in the actual package summary to submit package to Fedora Summary: Repository of Intel® Neural Compressor # Check if the automatically generated License and its spelling is correct for Fedora # https://docs.fedoraproject.org/en-US/packaging-guidelines/LicensingGuidelines/ License: Apache-2.0 URL: https://github.com/intel/neural-compressor Source: %{pypi_source neural_compressor} Source: https://raw.githubusercontent.com/intel/neural-compressor/a8cd9aa815ba7a94d9e0b4b028ce99eae22c2940/requirements.txt Source: https://raw.githubusercontent.com/intel/neural-compressor/a8cd9aa815ba7a94d9e0b4b028ce99eae22c2940/requirements_pt.txt Source: https://raw.githubusercontent.com/intel/neural-compressor/a8cd9aa815ba7a94d9e0b4b028ce99eae22c2940/requirements_tf.txt BuildArch: noarch BuildRequires: python3-devel BuildRequires: git # ONNX seems to be an optional dependency, but we need it because we want the import checks to succeed, and some parts use ONNX BuildRequires: python3-onnx BuildRequires: python3-onnxruntime # Torch also seems to be an optional dependency, but we are building with the pt extra BuildRequires: python3-torch BuildRequires: python3-transformers # Some code uses nltk, and it is packaged so require it BuildRequires: python3-nltk # Apparently some code also uses accelerate BuildRequires: python3-accelerate # Fill in the actual package description to submit package to Fedora %global _description %{expand: This is package 'neural-compressor' generated automatically by pyp2spec.} %description %_description %package -n python3-neural-compressor Summary: %{summary} %description -n python3-neural-compressor %_description # For official Fedora packages, review which extras should be actually packaged # See: https://docs.fedoraproject.org/en-US/packaging-guidelines/Python/#Extras %pyproject_extras_subpkg -n python3-neural-compressor pt %prep %autosetup -p1 -n neural_compressor-%{version} cp -a %{SOURCE1} requirements.txt cp -a %{SOURCE2} requirements_pt.txt cp -a %{SOURCE3} requirements_tf.txt # Fedora 41+ packages numpy 2.x.y and requirements.txt specifies numpy < 2, but # that, per this PR https://github.com/intel/neural-compressor/pull/1684 is only # because of old versions of scikit-learn, so remove the restriction! sed -i -e "s/numpy < 2.0/numpy/" requirements.txt sed -i -e "s/numpy<2.0/numpy/" neural_compressor.egg-info/requires.txt sed -i -e "s/numpy<2.0/numpy/" neural_compressor.egg-info/PKG-INFO sed -i -e "s/numpy<2.0/numpy/" PKG-INFO sed -i -e "s/numpy<2.0/numpy/" requirements_pt.txt # We also need to substitute the PyPI name for a special reduced opencv with # the full opencv that is packaged by Fedora sed -i -e "s/opencv-python-headless/opencv/" neural_compressor.egg-info/requires.txt sed -i -e "s/opencv-python-headless/opencv/" neural_compressor.egg-info/PKG-INFO sed -i -e "s/opencv-python-headless/opencv/" PKG-INFO sed -i -e "s/opencv-python-headless/opencv/" requirements.txt # Remove all the files that should only ship if we enable extras # This makes the import check pass # keras_utils, tf_utils needs tensorflow, which would be the +tf extra rm -r neural_compressor/adaptor/keras_utils rm -r neural_compressor/adaptor/tf_utils rm -r neural_compressor/tensorflow/ # needs xgboost rm neural_compressor/compression/hpo/search_algorithms.py rm neural_compressor/compression/hpo/__init__.py # needs bigcode_eval rm neural_compressor/evaluation/bigcode_eval/evaluator.py rm neural_compressor/evaluation/bigcode_eval/__init__.py # needs evaluate rm neural_compressor/evaluation/hf_eval/evaluator.py rm neural_compressor/evaluation/hf_eval/__init__.py # needs datasets rm neural_compressor/evaluation/hf_eval/hf_datasets/cnn_dailymail.py rm neural_compressor/transformers/quantization/utils.py rm neural_compressor/transformers/quantization/__init__.py rm neural_compressor/transformers/models/modeling_auto.py # TODO: I'm worries that some of these removals, like this one, break the package? rm neural_compressor/transformers/models/__init__.py # needs lm_eval rm neural_compressor/evaluation/lm_eval/accuracy.py rm neural_compressor/evaluation/lm_eval/models/huggingface.py rm neural_compressor/evaluation/lm_eval/__init__.py rm neural_compressor/evaluation/lm_eval/models/__init__.py # TODO: I'm worries that some of these removals, like this one, break the package? rm neural_compressor/transformers/__init__.py # needs habana_frameworks rm neural_compressor/torch/algorithms/fp8_quant/_quant_common/quant_config.py rm neural_compressor/torch/algorithms/fp8_quant/common.py rm neural_compressor/torch/algorithms/fp8_quant/__init__.py rm neural_compressor/torch/algorithms/fp8_quant/fp8_quant.py rm neural_compressor/torch/algorithms/fp8_quant/_core/measure.py rm neural_compressor/torch/algorithms/fp8_quant/prepare_quant/prepare_model.py rm neural_compressor/torch/algorithms/mixed_low_precision/custom_methods/gptq.py # needs ipex (intel_extensions_for_pytorch) rm neural_compressor/torch/algorithms/smooth_quant/utility.py rm neural_compressor/torch/algorithms/smooth_quant/__init__.py rm neural_compressor/torch/algorithms/smooth_quant/save_load.py rm neural_compressor/torch/algorithms/smooth_quant/smooth_quant.py rm neural_compressor/torch/algorithms/static_quant/static_quant.py rm neural_compressor/torch/algorithms/static_quant/__init__.py rm neural_compressor/torch/algorithms/static_quant/save_load.py # needs auto_round rm neural_compressor/torch/algorithms/weight_only/autoround.py # needs numba rm neural_compressor/torch/utils/bit_packer.py %generate_buildrequires # Keep only those extras which you actually want to package or use during tests %pyproject_buildrequires -x pt %build %pyproject_wheel %install %pyproject_install # Add top-level Python module names here as arguments, you can use globs %pyproject_save_files -l neural_compressor %check # Just drop the import checks so we can ship the package for now %pyproject_check_import %files -n python3-neural-compressor -f %{pyproject_files} %{_bindir}/incbench %changelog %autochangelog