From 476fb8e89242720a7cdd57400ba928de4dde9cc1 Mon Sep 17 00:00:00 2001 From: Jeffrey Morgan Date: Sat, 1 Jun 2024 19:24:33 -0700 Subject: [PATCH] Limit GPU lib search for now (#4777) * fix oneapi errors on windows 10 --- gpu/gpu.go | 32 +------------------------------- 1 file changed, 1 insertion(+), 31 deletions(-) diff --git a/gpu/gpu.go b/gpu/gpu.go index defdf04d..03e16702 100644 --- a/gpu/gpu.go +++ b/gpu/gpu.go @@ -16,13 +16,12 @@ import ( "os" "path/filepath" "runtime" - "strconv" "strings" "sync" "unsafe" - "github.com/ollama/ollama/format" "github.com/ollama/ollama/envconfig" + "github.com/ollama/ollama/format" ) type handles struct { @@ -105,8 +104,6 @@ func initGPUHandles() *handles { var cudartMgmtPatterns []string var nvcudaMgmtName string var nvcudaMgmtPatterns []string - var oneapiMgmtName string - var oneapiMgmtPatterns []string tmpDir, _ := PayloadsDir() switch runtime.GOOS { @@ -118,8 +115,6 @@ func initGPUHandles() *handles { // Aligned with driver, we can't carry as payloads nvcudaMgmtName = "nvcuda.dll" nvcudaMgmtPatterns = NvcudaWindowsGlobs - oneapiMgmtName = "ze_intel_gpu64.dll" - oneapiMgmtPatterns = OneapiWindowsGlobs case "linux": cudartMgmtName = "libcudart.so*" if tmpDir != "" { @@ -130,8 +125,6 @@ func initGPUHandles() *handles { // Aligned with driver, we can't carry as payloads nvcudaMgmtName = "libcuda.so*" nvcudaMgmtPatterns = NvcudaLinuxGlobs - oneapiMgmtName = "libze_intel_gpu.so" - oneapiMgmtPatterns = OneapiLinuxGlobs default: return gpuHandles } @@ -159,17 +152,6 @@ func initGPUHandles() *handles { } } - oneapiLibPaths := FindGPULibs(oneapiMgmtName, oneapiMgmtPatterns) - if len(oneapiLibPaths) > 0 { - deviceCount, oneapi, libPath := LoadOneapiMgmt(oneapiLibPaths) - if oneapi != nil { - slog.Debug("detected Intel GPUs", "library", libPath, "count", deviceCount) - gpuHandles.oneapi = oneapi - gpuHandles.deviceCount = deviceCount - return gpuHandles - } - } - return gpuHandles } @@ -245,18 +227,6 @@ func GetGPUInfo() GpuInfoList { // TODO potentially sort on our own algorithm instead of what the underlying GPU library does... resp = append(resp, gpuInfo) } - if gpuHandles.oneapi != nil { - gpuInfo := GpuInfo{ - Library: "oneapi", - } - C.oneapi_check_vram(*gpuHandles.oneapi, &memInfo) - var totalFreeMem float64 = float64(memInfo.free) * 0.95 // work-around: leave some reserve vram for mkl lib used in ggml-sycl backend. - memInfo.free = C.uint64_t(totalFreeMem) - gpuInfo.TotalMemory = uint64(memInfo.total) - gpuInfo.FreeMemory = uint64(memInfo.free) - gpuInfo.ID = strconv.Itoa(i) - resp = append(resp, gpuInfo) - } } // Then AMD