diff --git a/LLama/LLamaSharp.Runtime.targets b/LLama/LLamaSharp.Runtime.targets index 2523e55c1..76292aaf5 100644 --- a/LLama/LLamaSharp.Runtime.targets +++ b/LLama/LLamaSharp.Runtime.targets @@ -4,19 +4,19 @@ - + PreserveNewest runtimes/win-x64/native/noavx/llama.dll - + PreserveNewest runtimes/win-x64/native/noavx/ggml.dll - + PreserveNewest runtimes/win-x64/native/noavx/ggml-base.dll - + PreserveNewest runtimes/win-x64/native/noavx/ggml-cpu.dll @@ -130,19 +130,19 @@ - + PreserveNewest runtimes/linux-x64/native/noavx/libllama.so - + PreserveNewest runtimes/linux-x64/native/noavx/libggml.so - + PreserveNewest runtimes/linux-x64/native/noavx/libggml-base.so - + PreserveNewest runtimes/linux-x64/native/noavx/libggml-cpu.so @@ -338,7 +338,7 @@ runtimes/osx-x64/native/rosetta2/libllava_shared.dylib - + PreserveNewest runtimes/win-x64/native/noavx/llava_shared.dll @@ -367,7 +367,7 @@ runtimes/win-x64/native/vulkan/llava_shared.dll - + PreserveNewest runtimes/linux-x64/native/noavx/libllava_shared.so diff --git a/LLama/LLamaSharp.csproj b/LLama/LLamaSharp.csproj index 784f77221..b2d81711e 100644 --- a/LLama/LLamaSharp.csproj +++ b/LLama/LLamaSharp.csproj @@ -56,7 +56,7 @@ - 0827b2c1da-v5 + 0827b2c1da-v6 diff --git a/LLama/Native/Load/DefaultNativeLibrarySelectingPolicy.cs b/LLama/Native/Load/DefaultNativeLibrarySelectingPolicy.cs index 497902ba1..36ab0c0c8 100644 --- a/LLama/Native/Load/DefaultNativeLibrarySelectingPolicy.cs +++ b/LLama/Native/Load/DefaultNativeLibrarySelectingPolicy.cs @@ -22,12 +22,12 @@ public IEnumerable Apply(NativeLibraryConfig.Description descrip { if (description.UseCuda) { - yield return new NativeLibraryWithCuda(systemInfo.CudaMajorVersion, description.Library, description.SkipCheck); + yield return new NativeLibraryWithCuda(systemInfo.CudaMajorVersion, description.Library, description.AvxLevel, description.SkipCheck); } if (description.UseVulkan) { - yield return new NativeLibraryWithVulkan(systemInfo.VulkanVersion, description.Library, description.SkipCheck); + yield return new NativeLibraryWithVulkan(systemInfo.VulkanVersion, description.Library, description.AvxLevel, description.SkipCheck); } if((!description.UseCuda || !description.UseVulkan) || description.AllowFallback) diff --git a/LLama/Native/Load/NativeLibraryUtils.cs b/LLama/Native/Load/NativeLibraryUtils.cs index d0d853f63..13e68be4d 100644 --- a/LLama/Native/Load/NativeLibraryUtils.cs +++ b/LLama/Native/Load/NativeLibraryUtils.cs @@ -93,7 +93,7 @@ internal static IntPtr TryLoadLibrary(NativeLibraryConfig config, out INativeLib $"runtimes/{os}/native/{NativeLibraryConfig.AvxLevelToString(library.Metadata.AvxLevel)}", $"{libPrefix}ggml-cpu{ext}" )); - + // ggml-cuda if (library.Metadata.UseCuda) dependencyPaths.Add(Path.Combine(currentRuntimeDirectory, $"{libPrefix}ggml-cuda{ext}")); diff --git a/LLama/Native/Load/NativeLibraryWithCuda.cs b/LLama/Native/Load/NativeLibraryWithCuda.cs index 12da095dc..36dc4ca81 100644 --- a/LLama/Native/Load/NativeLibraryWithCuda.cs +++ b/LLama/Native/Load/NativeLibraryWithCuda.cs @@ -28,11 +28,13 @@ public NativeLibraryMetadata? Metadata /// /// /// + /// /// - public NativeLibraryWithCuda(int majorCudaVersion, NativeLibraryName libraryName, bool skipCheck) + public NativeLibraryWithCuda(int majorCudaVersion, NativeLibraryName libraryName, AvxLevel avxLevel, bool skipCheck) { _majorCudaVersion = majorCudaVersion; _libraryName = libraryName; + _avxLevel = avxLevel; _skipCheck = skipCheck; } diff --git a/LLama/Native/Load/NativeLibraryWithVulkan.cs b/LLama/Native/Load/NativeLibraryWithVulkan.cs index fe4eef01e..c3fe94de3 100644 --- a/LLama/Native/Load/NativeLibraryWithVulkan.cs +++ b/LLama/Native/Load/NativeLibraryWithVulkan.cs @@ -28,11 +28,13 @@ public NativeLibraryMetadata? Metadata /// /// /// + /// /// - public NativeLibraryWithVulkan(string? vulkanVersion, NativeLibraryName libraryName, bool skipCheck) + public NativeLibraryWithVulkan(string? vulkanVersion, NativeLibraryName libraryName, AvxLevel avxLevel, bool skipCheck) { _vulkanVersion = vulkanVersion; _libraryName = libraryName; + _avxLevel = avxLevel; _skipCheck = skipCheck; } diff --git a/LLama/runtimes/build/LLamaSharp.Backend.Cpu.nuspec b/LLama/runtimes/build/LLamaSharp.Backend.Cpu.nuspec index 382eb2ae8..debc99506 100644 --- a/LLama/runtimes/build/LLamaSharp.Backend.Cpu.nuspec +++ b/LLama/runtimes/build/LLamaSharp.Backend.Cpu.nuspec @@ -18,11 +18,11 @@ - - - - - + + + + + @@ -42,11 +42,11 @@ - - - - - + + + + +