IPEX v2.1.40+lnl

Customized intel_extension_for_pytorch wheel for Core Ultra Series 2 on windows.

• Built from IPEX source (intel@6c9c6a1)
• with oneAPI Base Toolkit 2024.2.1
• with Windows Arc Graphics Driver 32.0.101.5972 (the wheel should be compatible with any driver >= 32.0.101.5736)
• with build option USE_AOT_DEVLIST=xe2-lpg
• with python 3.11 (thus not compatible with other python versions)

Usage:

# libuv is required by intel torch
conda install libuv -y
# install intel torch/torchvision/torchaudio from intel channel
python -m pip install torch==2.1.0.post3 torchvision==0.16.0.post3 torchaudio==2.1.0.post3 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

# install IPEX for Core Ultra Series 2
python -m pip install intel_extension_for_pytorch-2.1.40+git6c9c6a1-cp311-cp311-win_amd64.whl
# install oneAPI dependencies
python -m pip install dpcpp-cpp-rt==2024.2.1 mkl-dpcpp==2024.2.1

This wheel should be (hopefully) compatible with IPEX-LLM 2.1.0.

IPEX v2.1.20+mtl-aot+oneapi

Customized torch, intel_extension_for_pytorch, torchvision, and torchaudio wheels for Core Ultra iGPU on Windows (compiled with driver 5084).

Please note the only AOT binaries contained in this release were built for Core Ultra iGPU, if you are using a Intel Arc A-Series dGPU, you probably want to use this AOT package for Arc dGPU, otherwise you may experience unaffordable JIT compilation overheads.

• The wheels were first built from source (4e792cc) with oneAPI DPCPP 2024.0, MKL 2024.0 and ocloc_win_101.5082, for AOT target mtl-p (Core Ultra iGPU).
• Then the torch and intel_extension_for_pytorch wheels were re-packaged with extra dll dependencies:
  • Extra dll files added to torch (location %PYTHON_ENV_DIR%\lib\site-packages\torch\lib)
    • uv.dll from libuv
    • libiomp5md.dll from oneAPI DPCPP compiler 2024.0
  • Extra dll files added to intel_extension_for_pytorch (location %PYTHON_ENV_DIR%\lib\site-packages\intel_extension_for_pytorch\bin)
    • sycl7.dll;pi_level_zero.dll;pi_win_proxy_loader.dll;mkl_core.2.dll;mkl_sycl_blas.4.dll;mkl_sycl_lapack.4.dll;mkl_sycl_dft.4.dll;mkl_tbb_thread.2.dll;libmmd.dll;svml_dispmd.dll from oneAPI DPCPP 2024.0 and MKL 2024.0

With all the dependent dll files baked in torch and intel_extension_for_pytorch wheels, the users can simply install and use IPEX without installing extra oneAPI packages.

IPEX v2.1.10+xpu+oneapi

The torch and intel_extension_for_pytorch wheels were re-packaged with extra dll dependencies on top of official IPEX v2.1.10+xpu release (Intel® Data Center GPU Flex Series, Intel® Data Center GPU Max Series and Intel® Arc™ A-Series Graphics):

• Extra dll files added to torch (location %PYTHON_ENV_DIR%\lib\site-packages\torch\lib)
  • uv.dll from libuv 1.44
  • libiomp5md.dll from oneAPI DPCPP compiler 2024.0
• Extra dll files added to intel_extension_for_pytorch (location %PYTHON_ENV_DIR%\lib\site-packages\intel_extension_for_pytorch\bin)
  • sycl7.dll;pi_level_zero.dll;pi_win_proxy_loader.dll;mkl_core.2.dll;mkl_sycl_blas.4.dll;mkl_sycl_lapack.4.dll;mkl_sycl_dft.4.dll;mkl_tbb_thread.2.dll;libmmd.dll;svml_dispmd.dll from oneAPI DPCPP 2024.0 and MKL 2024.0

With all the dependent dll files baked in torch and intel_extension_for_pytorch wheels, the users can simply install and use IPEX without installing extra oneAPI packages.

IPEX v2.0.110+mtl-aot+oneapi

Customized torch, intel_extension_for_pytorch and torchvision wheels for Core Ultra iGPU on Windows (compiled with driver 5084).

Please note the only AOT binaries contained in this release were built for Core Ultra iGPU, if you are using a Intel Arc A-Series dGPU, you probably want to use this AOT package for Arc dGPU, otherwise you may experience unaffordable JIT compilation overheads.

• The wheels were first built from source (632f70a) with oneAPI DPCPP 2023.2, MKL 2023.2 and ocloc_win_101.5082, for AOT target mtl-p (Core Ultra iGPU).
• Then the torch and intel_extension_for_pytorch wheels were re-packaged with extra dll dependencies:
  • Extra dll files added to torch (location %PYTHON_ENV_DIR%\lib\site-packages\torch\lib)
    • uv.dll from libuv 1.39
    • libiomp5md.dll from oneAPI DPCPP compiler 2023.2.1
  • Extra dll files added to intel_extension_for_pytorch (location %PYTHON_ENV_DIR%\lib\site-packages\intel_extension_for_pytorch\bin)
    • sycl6.dll, pi_level_zero.dll, pi_win_proxy_loader.dll, mkl_core.2.dll, mkl_sycl.3.dll, mkl_tbb_thread.2.dll, libmmd.dll, svml_dispmd.dll from oneAPI DPCPP 2023.2 and MKL 2023.2

With all the dependent dll files baked in torch and intel_extension_for_pytorch wheels, the users can simply install and use IPEX without installing extra oneAPI packages.

IPEX xpu-master AOT (Arc) with oneAPI dll bundle

Customized torch, intel_extension_for_pytorch and torchvision wheels for Intel Arc GPU on Windows (compiled with driver 4676).

• The wheels were first built from source (ipex xpu-master branch) with oneAPI DPCPP 2023.2 and MKL 2023.2, for AOT target ats-m150 (Arc Series).
• Then the torch and intel_extension_for_pytorch wheels were re-packaged with extra dll dependencies:
  • Extra dll files added to torch (location %PYTHON_ENV_DIR%\lib\site-packages\torch\lib)
    • uv.dll from libuv 1.39
    • libiomp5md.dll from oneAPI DPCPP compiler 2023.2.1
  • Extra dll files added to intel_extension_for_pytorch (location %PYTHON_ENV_DIR%\lib\site-packages\intel_extension_for_pytorch\bin)
    • sycl6.dll, pi_level_zero.dll, pi_win_proxy_loader.dll, mkl_core.2.dll, mkl_sycl.3.dll, mkl_tbb_thread.2.dll, libmmd.dll, svml_dispmd.dll from oneAPI DPCPP 2023.2 and MKL 2023.2

With all the dependent dll files baked in torch and intel_extension_for_pytorch wheels, the users can simply install and use IPEX without installing extra oneAPI packages.

IPEX xpu-master AOT (ats-m150) win64 wheel (built from source)

Revert "Enable quant save/load through prepack fn registration (#3078)"

This reverts commit 7db2f1c47495e6a26cfd212b191eb8a66524096e.