Is it possible to have a SYCL backend runtime.
It's really faster on Intel GPU than the vulkan backend. When using a llama.cpp Sycl version it's 2x to 3x faster than the vulkan version.
Or can I build my own runtime using a llama.cpp sycl version ( using ggml.dll) ?
Is it possible to have a SYCL backend runtime. It's really faster on Intel GPU than the vulkan backend. When using a llama.cpp Sycl version it's 2x to 3x faster than the vulkan version.
Or can I build my own runtime using a llama.cpp sycl version ( using ggml.dll) ?
Here the doc to build the backend : https://github.com/ggerganov/llama.cpp/blob/master/docs/backend/SYCL.md