windows CUDA support (#1262)

* Support cuda build in Windows
* Enable dynamic NumGPU allocation for Windows

.gitignore

@@ -7,3 +7,4 @@ dist
 ollama
 ggml-metal.metal
 .cache
+*.exe

llm/llama.cpp/generate_windows.go

@@ -14,3 +14,11 @@ package llm
 //go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
 //go:generate cmake --build gguf/build/cpu --target server --config Release
 //go:generate cmd /c move gguf\build\cpu\bin\Release\server.exe gguf\build\cpu\bin\Release\ollama-runner.exe
+
+//go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
+//go:generate cmake --build ggml/build/cuda --target server --config Release
+//go:generate cmd /c move ggml\build\cuda\bin\Release\server.exe ggml\build\cuda\bin\Release\ollama-runner.exe
+
+//go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
+//go:generate cmake --build gguf/build/cuda --target server --config Release
+//go:generate cmd /c move gguf\build\cuda\bin\Release\server.exe gguf\build\cuda\bin\Release\ollama-runner.exe

llm/llama.go

@@ -84,6 +84,7 @@ func chooseRunners(workDir, runnerType string) []ModelRunner {
 	case "windows":
 		// TODO: select windows GPU runner here when available
 		runners = []ModelRunner{
+			{Path: path.Join(buildPath, "cuda", "bin", "Release", "ollama-runner.exe"), Accelerated: true},
 			{Path: path.Join(buildPath, "cpu", "bin", "Release", "ollama-runner.exe")},
 		}
 	default:
@@ -269,7 +270,7 @@ func NumGPU(numLayer, fileSizeBytes int64, opts api.Options) int {
 	if opts.NumGPU != -1 {
 		return opts.NumGPU
 	}
-	if runtime.GOOS == "linux" {
+	if runtime.GOOS == "linux" || runtime.GOOS == "windows" {
 		freeBytes, err := CheckVRAM()
 		if err != nil {
 			if !errors.Is(err, errNvidiaSMI) {