leave some vram for mkl lib used in ggml-sycl backend

Dockerfile

@@ -50,7 +50,7 @@ RUN mkdir /tmp/scratch && \
     (cd /tmp/scratch/ && tar czvf /go/src/github.com/ollama/ollama/dist/deps/ollama-linux-amd64-rocm.tgz . )
 
 
-FROM --platform=linux/amd64 intel/oneapi-basekit:2024.0.1-devel-rockylinux9 AS oneapi-build-amd64
+FROM --platform=linux/amd64 intel/oneapi-basekit:2024.1.1-devel-rockylinux9 AS oneapi-build-amd64
 ARG CMAKE_VERSION
 COPY ./scripts/rh_linux_deps.sh /
 RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
@@ -147,7 +147,7 @@ CMD ["serve"]
 # e.g. docker run -it  -v "$(pwd):/app:Z" --device /dev/dri/renderD128:/dev/dri/renderD129 --device /dev/dri/card1:/dev/dri/card1 IMAGE_ID
 # host machine need to install Intel GPU driver correctly.
 # to detect the device info which you want to mount, run sudo intel_gpu_top -L
-FROM --platform=linux/amd64 intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04 as runtime-oneapi
+FROM --platform=linux/amd64 intel/oneapi-basekit:2024.1.1-devel-ubuntu22.04 as runtime-oneapi
 RUN update-pciids
 COPY --from=build-amd64 /go/src/github.com/ollama/ollama/ollama /bin/ollama
 EXPOSE 11434

gpu/gpu.go

@@ -253,6 +253,8 @@ func GetGPUInfo() GpuInfo {
 					val = "DeviceType:gpu"
 					os.Setenv("SYCL_DEVICE_ALLOWLIST", val)
 				}
+				var totalFreeMem float64 = float64(memInfo.free) * 0.9 // leave some reverse vram for mkl lib used in ggml-sycl backend.
+				memInfo.free = C.uint64_t(totalFreeMem)
 				slog.Info(fmt.Sprintf("oneAPI integrated GPU detected - SYCL_DEVICE_ALLOWLIST=%s", val))
 			}
 			resp.Library = "oneapi"