Is there a release version that supports Intel ARC A770 graphics card acceleration?

[simplete]

I manually compiled the SYCL backend successfully, but encountered an error when running it:

Executing the test code as follows:

There is an issue with the compiled DLL file: Tried replacing it with the DLL from OneAPI, and the DLL dependency issue seems to be resolved, but executing the test code still reports an error: "C:\Program Files (x86)\Intel\oneAPI\dnnl\latest\bin\dnnl.dll" & "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\bin\sycl8.dll"==>D:\miniconda\envs\comfyui\Lib\site-packages\llama_cpp\lib "C:\Program Files (x86)\Intel\oneAPI\dnnl\latest\bin\dnnl.dll" & "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\bin\sycl8.dll"==>D:\miniconda\envs\comfyui\Lib\site-packages\llama_cpp\bin I don't know where the problem is. The compilation environment is as follows: Python 3.11 OneAPI 2025.3 Visual Studio version 2026 18.3.2 Compilation script: conda activate comfyui call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" set "ONEAPI_ROOT=C:\Program Files (x86)\Intel\oneAPI" set "PATH=%PATH%;D:\llama-cpp-ipex-llm-2.2.0-win" set "CPATH=%CPATH%;%ONEAPI_ROOT%\compiler\latest\windows\include" set "LIB=%LIB%;%ONEAPI_ROOT%\compiler\latest\windows\lib" cd /d D:\llama-cpp-python-main set CMAKE_ARGS=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx set FORCE_CMAKE=1 pip wheel . --no-deps --verbose --no-cache-dir

[simplete]

llama_cpp_python-0.3.30-cp311-cp311-win_amd64.z03.zip
llama_cpp_python-0.3.30-cp311-cp311-win_amd64.zip
llama_cpp_python-0.3.30-cp311-cp311-win_amd64.z01.zip
llama_cpp_python-0.3.30-cp311-cp311-win_amd64.z02.zip
This is the result of my compilation.

[JamePeng]

The above message clearly indicates that the ggml library cannot be found. This likely means the compilation failed or the compiled library is incompatible.

The two paths above can also be provided as screenshots to show what files are present after successful installation.

I consistently recommend using the Vulkan backend for Intel and AMD graphics cards to achieve more stable compilation results. The Vulkan branch is optimized more frequently and proactively, resulting in better compatibility and stability.

[simplete]

Switched to Vulkan backend, compiled successfully on the first try, much easier to use, very stable, and no additional steps required..
Thanks very much!!!
Vulkan backend.log

[simplete]

This is driving me crazy. I had it working fine in the afternoon—using the SYCL backend, it was fast, around 15 seconds—but after lunch, it just won't work at all.
It works by replacing the DLL file released at https://github.com/ggml-org/llama.cpp/releases.