Merge pull request #3850 from dhiltgen/windows_packaging

Move nested payloads to installer and zip file on windows

.github/workflows/release.yaml

@@ -103,6 +103,7 @@ jobs:
           path: |
             llm/build/**/bin/*
             llm/build/**/*.a
+            dist/windows-amd64/**
 
   # ROCm generation step
   generate-windows-rocm:
@@ -173,7 +174,9 @@ jobs:
       - uses: actions/upload-artifact@v4
         with:
           name: generate-windows-rocm
-          path: llm/build/**/bin/*
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
       - uses: actions/upload-artifact@v4
         with:
           name: windows-rocm-deps
@@ -253,7 +256,9 @@ jobs:
       - uses: actions/upload-artifact@v4
         with:
           name: generate-windows-cuda
-          path: llm/build/**/bin/*
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
       - uses: actions/upload-artifact@v4
         with:
           name: windows-cuda-deps
@@ -306,11 +311,15 @@ jobs:
       - uses: actions/download-artifact@v4
         with:
           name: generate-windows-cpu
-          path: llm/build
+          path: |
+            llm/build
+            dist/windows-amd64
       - uses: actions/download-artifact@v4
         with:
           name: generate-windows-cuda
-          path: llm/build
+          path: |
+            llm/build
+            dist/windows-amd64
       - uses: actions/download-artifact@v4
         with:
           name: windows-cuda-deps
@@ -322,7 +331,9 @@ jobs:
       - uses: actions/download-artifact@v4
         with:
           name: generate-windows-rocm
-          path: llm/build
+          path: |
+            llm/build
+            dist/windows-amd64
       - run: dir llm/build
       - run: |
           $gopath=(get-command go).source | split-path -parent
@@ -337,7 +348,9 @@ jobs:
       - uses: actions/upload-artifact@v4
         with:
           name: dist-windows
-          path: dist/*.exe
+          path: |
+            dist/OllamaSetup.exe
+            dist/ollama-windows-*.zip
 
   # Linux x86 assets built using the container based build
   build-linux-amd64:

.github/workflows/test.yaml

@@ -103,7 +103,9 @@ jobs:
       - uses: actions/upload-artifact@v4
         with:
           name: cuda-${{ matrix.cuda-version }}-libraries
-          path: llm/build/**/bin/*
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
   generate-rocm:
     needs: [changes]
     if: ${{ needs.changes.outputs.GENERATE_ROCM == 'True' }}
@@ -134,7 +136,9 @@ jobs:
       - uses: actions/upload-artifact@v4
         with:
           name: rocm-${{ matrix.rocm-version }}-libraries
-          path: llm/build/**/bin/*
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
 
   # ROCm generation step
   generate-windows-rocm:
@@ -253,11 +257,6 @@ jobs:
           mkdir -p llm/build/darwin/$ARCH/stub/bin
           touch llm/build/darwin/$ARCH/stub/bin/ollama_llama_server
         if: ${{ startsWith(matrix.os, 'macos-') }}
-      - run: |
-          mkdir -p llm/build/windows/$ARCH/stub/bin
-          touch llm/build/windows/$ARCH/stub/bin/ollama_llama_server
-        if: ${{ startsWith(matrix.os, 'windows-') }}
-        shell: bash
       - uses: golangci/golangci-lint-action@v4
         with:
           args: --timeout 8m0s -v
@@ -299,10 +298,6 @@ jobs:
           mkdir -p llm/build/darwin/$ARCH/stub/bin
           touch llm/build/darwin/$ARCH/stub/bin/ollama_llama_server
         if: ${{ startsWith(matrix.os, 'macos-') }}
-      - run: |
-          mkdir -p llm/build/windows/$ARCH/stub/bin
-          touch llm/build/windows/$ARCH/stub/bin/ollama_llama_server
-        if: ${{ startsWith(matrix.os, 'windows-') }}
         shell: bash
       - run: go generate ./...
       - run: go build

app/ollama.iss

@@ -88,7 +88,8 @@ DialogFontSize=12
 [Files]
 Source: ".\app.exe"; DestDir: "{app}"; DestName: "{#MyAppExeName}" ; Flags: ignoreversion 64bit
 Source: "..\ollama.exe"; DestDir: "{app}"; Flags: ignoreversion 64bit
-Source: "..\dist\windeps\*.dll"; DestDir: "{app}"; Flags: ignoreversion 64bit
+Source: "..\dist\windows-amd64\*.dll"; DestDir: "{app}"; Flags: ignoreversion 64bit
+Source: "..\dist\windows-amd64\ollama_runners\*"; DestDir: "{app}\ollama_runners"; Flags: ignoreversion 64bit recursesubdirs
 Source: "..\dist\ollama_welcome.ps1"; DestDir: "{app}"; Flags: ignoreversion
 Source: ".\assets\app.ico"; DestDir: "{app}"; Flags: ignoreversion
 ; Assumes v5.7, may need adjustments for v6

gpu/assets.go

@@ -24,6 +24,35 @@ func PayloadsDir() (string, error) {
 	defer lock.Unlock()
 	var err error
 	if payloadsDir == "" {
+		runnersDir := os.Getenv("OLLAMA_RUNNERS_DIR")
+		// On Windows we do not carry the payloads inside the main executable
+		if runtime.GOOS == "windows" && runnersDir == "" {
+			appExe, err := os.Executable()
+			if err != nil {
+				slog.Error("failed to lookup executable path", "error", err)
+				return "", err
+			}
+			// Try a few variations to improve developer experience when building from source in the local tree
+			for _, d := range []string{".", "windows-" + runtime.GOARCH, "dist\\windows-" + runtime.GOARCH} {
+				candidate := filepath.Join(filepath.Dir(appExe), d, "ollama_runners")
+				_, err := os.Stat(candidate)
+				if err == nil {
+					runnersDir = candidate
+					break
+				}
+			}
+			if runnersDir == "" {
+				err = fmt.Errorf("unable to locate llm runner directory.  Set OLLAMA_RUNNERS_DIR to the location of 'ollama_runners'")
+				slog.Error("incomplete distribution", "error", err)
+				return "", err
+			}
+		}
+		if runnersDir != "" {
+			payloadsDir = runnersDir
+			return payloadsDir, nil
+		}
+
+		// The remainder only applies on non-windows where we still carry payloads in the main executable
 		cleanupTmpDirs()
 		tmpDir := os.Getenv("OLLAMA_TMPDIR")
 		if tmpDir == "" {
@@ -88,7 +117,8 @@ func cleanupTmpDirs() {
 func Cleanup() {
 	lock.Lock()
 	defer lock.Unlock()
-	if payloadsDir != "" {
+	runnersDir := os.Getenv("OLLAMA_RUNNERS_DIR")
+	if payloadsDir != "" && runnersDir == "" && runtime.GOOS != "windows" {
 		// We want to fully clean up the tmpdir parent of the payloads dir
 		tmpDir := filepath.Clean(filepath.Join(payloadsDir, ".."))
 		slog.Debug("cleaning up", "dir", tmpDir)

llm/generate/gen_windows.ps1

@@ -35,6 +35,7 @@ function init_vars {
         )
     $script:cmakeTargets = @("ollama_llama_server")
     $script:ARCH = "amd64" # arm not yet supported.
+    $script:DIST_BASE = "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_runners"
     if ($env:CGO_CFLAGS -contains "-g") {
         $script:cmakeDefs += @("-DCMAKE_VERBOSE_MAKEFILE=on", "-DLLAMA_SERVER_VERBOSE=on", "-DCMAKE_BUILD_TYPE=RelWithDebInfo")
         $script:config = "RelWithDebInfo"
@@ -55,7 +56,6 @@ function init_vars {
     } else {
         $script:CUDA_LIB_DIR=$env:CUDA_LIB_DIR
     }
-    $script:GZIP=(get-command -ea 'silentlycontinue' gzip).path
     $script:DUMPBIN=(get-command -ea 'silentlycontinue' dumpbin).path
     if ($null -eq $env:CMAKE_CUDA_ARCHITECTURES) {
         $script:CMAKE_CUDA_ARCHITECTURES="50;52;61;70;75;80"
@@ -134,21 +134,18 @@ function sign {
     }
 }
 
-function compress {
+function install {
-    if ($script:GZIP -eq $null) {
+    write-host "Installing binaries to dist dir ${script:distDir}"
-        write-host "gzip not installed, not compressing files"
+    mkdir ${script:distDir} -ErrorAction SilentlyContinue
-        return
-    }
-    write-host "Compressing binaries..."
     $binaries = dir "${script:buildDir}/bin/*.exe"
     foreach ($file in $binaries) {
-        & "$script:GZIP" --best -f $file
+        copy-item -Path $file -Destination ${script:distDir} -Force
     }
 
-    write-host "Compressing dlls..."
+    write-host "Installing dlls to dist dir ${script:distDir}"
     $dlls = dir "${script:buildDir}/bin/*.dll"
     foreach ($file in $dlls) {
-        & "$script:GZIP" --best -f $file
+        copy-item -Path $file -Destination ${script:distDir} -Force
     }
 }
 
@@ -209,26 +206,29 @@ build
     init_vars
     $script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DLLAMA_AVX=off", "-DLLAMA_AVX2=off", "-DLLAMA_AVX512=off", "-DLLAMA_FMA=off", "-DLLAMA_F16C=off") + $script:cmakeDefs
     $script:buildDir="../build/windows/${script:ARCH}/cpu"
+    $script:distDir="$script:DIST_BASE\cpu"
     write-host "Building LCD CPU"
     build
     sign
-    compress
+    install
 
     init_vars
     $script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DLLAMA_AVX=on", "-DLLAMA_AVX2=off", "-DLLAMA_AVX512=off", "-DLLAMA_FMA=off", "-DLLAMA_F16C=off") + $script:cmakeDefs
     $script:buildDir="../build/windows/${script:ARCH}/cpu_avx"
+    $script:distDir="$script:DIST_BASE\cpu_avx"
     write-host "Building AVX CPU"
     build
     sign
-    compress
+    install
 
     init_vars
     $script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DLLAMA_AVX=on", "-DLLAMA_AVX2=on", "-DLLAMA_AVX512=off", "-DLLAMA_FMA=on", "-DLLAMA_F16C=on") + $script:cmakeDefs
     $script:buildDir="../build/windows/${script:ARCH}/cpu_avx2"
+    $script:distDir="$script:DIST_BASE\cpu_avx2"
     write-host "Building AVX2 CPU"
     build
     sign
-    compress
+    install
 } else {
     write-host "Skipping CPU generation step as requested"
 }
@@ -242,6 +242,7 @@ if ($null -ne $script:CUDA_LIB_DIR) {
     }
     init_vars
     $script:buildDir="../build/windows/${script:ARCH}/cuda$script:CUDA_VARIANT"
+    $script:distDir="$script:DIST_BASE\cuda$script:CUDA_VARIANT"
     $script:cmakeDefs += @("-A", "x64", "-DLLAMA_CUDA=ON", "-DLLAMA_AVX=on", "-DLLAMA_AVX2=off", "-DCUDAToolkit_INCLUDE_DIR=$script:CUDA_INCLUDE_DIR", "-DCMAKE_CUDA_ARCHITECTURES=${script:CMAKE_CUDA_ARCHITECTURES}")
     if ($null -ne $env:OLLAMA_CUSTOM_CUDA_DEFS) {
         write-host "OLLAMA_CUSTOM_CUDA_DEFS=`"${env:OLLAMA_CUSTOM_CUDA_DEFS}`""
@@ -250,7 +251,7 @@ if ($null -ne $script:CUDA_LIB_DIR) {
     }
     build
     sign
-    compress
+    install
 }
 
 if ($null -ne $env:HIP_PATH) {
@@ -261,6 +262,7 @@ if ($null -ne $env:HIP_PATH) {
 
     init_vars
     $script:buildDir="../build/windows/${script:ARCH}/rocm$script:ROCM_VARIANT"
+    $script:distDir="$script:DIST_BASE\rocm$script:ROCM_VARIANT"
     $script:cmakeDefs += @(
         "-G", "Ninja", 
         "-DCMAKE_C_COMPILER=clang.exe",
@@ -292,9 +294,9 @@ if ($null -ne $env:HIP_PATH) {
         & "$script:DUMPBIN" /dependents "${script:buildDir}/bin/ollama_llama_server.exe" | select-string ".dll"
     }
     sign
-    compress
+    install
 }
 
 
 cleanup
-write-host "`ngo generate completed.  LLM runners: $(get-childitem -path ${script:SRC_DIR}\llm\build\windows\${script:ARCH})"
+write-host "`ngo generate completed.  LLM runners: $(get-childitem -path $script:DIST_BASE)"

llm/llm_windows.go

@@ -2,5 +2,5 @@ package llm
 
 import "embed"
 
-//go:embed build/windows/*/*/bin/*
+// unused on windows
 var libEmbed embed.FS

llm/payload.go

@@ -26,13 +26,15 @@ func Init() error {
 		return err
 	}
 
-	slog.Info("extracting embedded files", "dir", payloadsDir)
+	if runtime.GOOS != "windows" {
-	binGlob := "build/*/*/*/bin/*"
+		slog.Info("extracting embedded files", "dir", payloadsDir)
+		binGlob := "build/*/*/*/bin/*"
 
-	// extract server libraries
+		// extract server libraries
-	err = extractFiles(payloadsDir, binGlob)
+		err = extractFiles(payloadsDir, binGlob)
-	if err != nil {
+		if err != nil {
-		return fmt.Errorf("extract binaries: %v", err)
+			return fmt.Errorf("extract binaries: %v", err)
+		}
 	}
 
 	var variants []string

scripts/build_windows.ps1

@@ -30,7 +30,7 @@ function checkEnv() {
     
     $script:INNO_SETUP_DIR=(get-item "C:\Program Files*\Inno Setup*\")[0]
 
-    $script:DEPS_DIR="${script:SRC_DIR}\dist\windeps"
+    $script:DEPS_DIR="${script:SRC_DIR}\dist\windows-amd64"
     $env:CGO_ENABLED="1"
     echo "Checking version"
     if (!$env:VERSION) {
@@ -81,8 +81,8 @@ function buildOllama() {
             /csp "Google Cloud KMS Provider" /kc ${env:KEY_CONTAINER} ollama.exe
         if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
     }
-    New-Item -ItemType Directory -Path .\dist -Force
+    New-Item -ItemType Directory -Path .\dist\windows-amd64\ -Force
-    cp .\ollama.exe .\dist\ollama-windows-amd64.exe
+    cp .\ollama.exe .\dist\windows-amd64\ollama-windows-amd64.exe
 }
 
 function buildApp() {
@@ -101,7 +101,6 @@ function buildApp() {
 function gatherDependencies() {
     write-host "Gathering runtime dependencies"
     cd "${script:SRC_DIR}"
-    rm -ea 0 -recurse -force -path "${script:DEPS_DIR}"
     md "${script:DEPS_DIR}" -ea 0 > $null
 
     # TODO - this varies based on host build system and MSVC version - drive from dumpbin output
@@ -124,7 +123,15 @@ function gatherDependencies() {
             if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
         }
     }
-
+    if ($null -ne $env:HIP_PATH) {
+        # Assumes v5.7, may need adjustments for v6
+        rm -ea 0 -recurse -force -path "${script:DEPS_DIR}\rocm\"
+        md "${script:DEPS_DIR}\rocm\rocblas\library\" -ea 0 > $null
+        cp "${env:HIP_PATH}\bin\hipblas.dll" "${script:DEPS_DIR}\rocm\"
+        cp "${env:HIP_PATH}\bin\rocblas.dll" "${script:DEPS_DIR}\rocm\"
+        # amdhip64.dll dependency comes from the driver and must be installed on the host to use AMD GPUs
+        cp "${env:HIP_PATH}\bin\rocblas\library\*" "${script:DEPS_DIR}\rocm\rocblas\library\"
+    }
 }
 
 function buildInstaller() {
@@ -139,12 +146,18 @@ function buildInstaller() {
     if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
 }
 
+function distZip() {
+    write-host "Generating stand-alone distribution zip file ${script:SRC_DIR}\dist\ollama-windows-amd64.zip"
+    Compress-Archive -Path "${script:SRC_DIR}\dist\windows-amd64\*" -DestinationPath "${script:SRC_DIR}\dist\ollama-windows-amd64.zip" -Force
+}
+
 try {
     checkEnv
     buildOllama
     buildApp
     gatherDependencies
     buildInstaller
+    distZip
 } catch {
     write-host "Build Failed"
     write-host $_

server/sched_test.go

@@ -92,6 +92,7 @@ func TestLoad(t *testing.T) {
 	runner := s.loaded["dummy_model_path"]
 	require.NotNil(t, runner)
 	require.Equal(t, uint(0), runner.refCount)
+	time.Sleep(1 * time.Millisecond)
 	require.Len(t, s.expiredCh, 1)
 }