GPT-QModel v7.3.2

What's Changed

• Bump actions/setup-python from 6 to 7 in the github-actions group by @dependabot[bot] in #2957
• fix: support Transformers 5.14 cache and dtype APIs by @Qubitium in #2959
• Add Laguna S 2.1 model support by @Qubitium in #2960
• Fix Laguna Evalution runtime configuration by @Qubitium in #2961
• Update Evalution dependency version to 0.0.9 by @Qubitium in #2962
• Clamp reconstructed quantization codes before packing by @Qubitium in #2963
• Fix unsafe EoRA wq aliasing by @Qubitium in #2965
• [MODEL] support inkling by @ZX-ModelCloud in #2958
• fix(logger): stop _AdaptiveLoggerProxy.getattr from recursing on deepcopy (fixes #2966) by @Anai-Guo in #2968
• fix: fence parallel calibration workers before completion by @Qubitium in #2969
• [MODEL]support intern_s2_preview by @ZX-ModelCloud in #2964
• Support quantized embedding inference without backend misselection by @Qubitium in #2971
• [MODEL] support solar_open2 by @ZX-ModelCloud in #2970
• [MODEL] support solar_open by @ZX-ModelCloud in #2972
• fix: per-module dynamic kernel selection for mixed-bitwidth models by @Qubitium in #2973
• fix(cpp/tests): C++20 JIT flags for PyTorch 2.14 nightly by @Qubitium in #2974
• deps: bump logbar and pypcre to latest PyPI releases by @Qubitium in #2975
• test(llama3_2): update fast-mode MARLIN baseline scores by @Qubitium in #2976
• Update version.py by @Qubitium in #2977

Full Changelog: v7.3.1...v7.3.2