handle intermediate blobs

- FROM /path/to/{safetensors,pytorch}
- FROM /path/to/fp{16,32}.bin
- FROM model:fp{16,32}

.gitignore

@@ -12,3 +12,4 @@ ggml-metal.metal
 test_data
 *.crt
 llm/build
+__debug_bin*

README.md

@@ -1,5 +1,5 @@
 <div align="center">
-  <img alt="ollama" height="200px" src="https://github.com/ollama/ollama/assets/3325447/0d0b44e2-8f4a-4e99-9b52-a5c1c741c8f7">
+ <img alt="ollama" height="200px" src="https://github.com/ollama/ollama/assets/3325447/0d0b44e2-8f4a-4e99-9b52-a5c1c741c8f7">
 </div>
 
 # Ollama
@@ -51,7 +51,7 @@ Here are some example models that can be downloaded:
 | ------------------ | ---------- | ----- | ------------------------------ |
 | Llama 3            | 8B         | 4.7GB | `ollama run llama3`            |
 | Llama 3            | 70B        | 40GB  | `ollama run llama3:70b`        |
-| Phi-3              | 3,8B       | 2.3GB | `ollama run phi3`              |
+| Phi-3              | 3.8B       | 2.3GB | `ollama run phi3`              |
 | Mistral            | 7B         | 4.1GB | `ollama run mistral`           |
 | Neural Chat        | 7B         | 4.1GB | `ollama run neural-chat`       |
 | Starling           | 7B         | 4.1GB | `ollama run starling-lm`       |
@@ -173,7 +173,7 @@ I'm a basic program that prints the famous "Hello, world!" message to the consol
 The image features a yellow smiley face, which is likely the central focus of the picture.
 ```
 
-### Pass in prompt as arguments
+### Pass the prompt as an argument
 
 ```
 $ ollama run llama3 "Summarize this file: $(cat README.md)"
@@ -294,7 +294,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [RAGFlow: Open-source Retrieval-Augmented Generation engine based on deep document understanding](https://github.com/infiniflow/ragflow)
 - [chat: chat web app for teams](https://github.com/swuecho/chat)
 - [Lobe Chat](https://github.com/lobehub/lobe-chat) with [Integrating Doc](https://lobehub.com/docs/self-hosting/examples/ollama)
-- [Ollama RAG Chatbot: Local Chat with multiples PDFs using Ollama and RAG.](https://github.com/datvodinh/rag-chatbot.git)
+- [Ollama RAG Chatbot: Local Chat with multiple PDFs using Ollama and RAG.](https://github.com/datvodinh/rag-chatbot.git)
 
 ### Terminal
 
@@ -384,4 +384,4 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Discord-Ollama Chat Bot](https://github.com/kevinthedang/discord-ollama) (Generalized TypeScript Discord Bot w/ Tuning Documentation)
 
 ### Supported backends 
-- [llama.cpp](https://github.com/ggerganov/llama.cpp) project founded by Georgi Gerganov. 
+- [llama.cpp](https://github.com/ggerganov/llama.cpp) project founded by Georgi Gerganov. 

api/client.go

@@ -18,6 +18,7 @@ import (
 	"net/url"
 	"os"
 	"runtime"
+	"strconv"
 	"strings"
 
 	"github.com/ollama/ollama/format"
@@ -57,12 +58,36 @@ func checkError(resp *http.Response, body []byte) error {
 // If the variable is not specified, a default ollama host and port will be
 // used.
 func ClientFromEnvironment() (*Client, error) {
+	ollamaHost, err := GetOllamaHost()
+	if err != nil {
+		return nil, err
+	}
+
+	return &Client{
+		base: &url.URL{
+			Scheme: ollamaHost.Scheme,
+			Host:   net.JoinHostPort(ollamaHost.Host, ollamaHost.Port),
+		},
+		http: http.DefaultClient,
+	}, nil
+}
+
+type OllamaHost struct {
+	Scheme string
+	Host   string
+	Port   string
+}
+
+func GetOllamaHost() (OllamaHost, error) {
 	defaultPort := "11434"
 
-	scheme, hostport, ok := strings.Cut(os.Getenv("OLLAMA_HOST"), "://")
+	hostVar := os.Getenv("OLLAMA_HOST")
+	hostVar = strings.TrimSpace(strings.Trim(strings.TrimSpace(hostVar), "\"'"))
+
+	scheme, hostport, ok := strings.Cut(hostVar, "://")
 	switch {
 	case !ok:
-		scheme, hostport = "http", os.Getenv("OLLAMA_HOST")
+		scheme, hostport = "http", hostVar
 	case scheme == "http":
 		defaultPort = "80"
 	case scheme == "https":
@@ -82,12 +107,14 @@ func ClientFromEnvironment() (*Client, error) {
 		}
 	}
 
-	return &Client{
-		base: &url.URL{
-			Scheme: scheme,
-			Host:   net.JoinHostPort(host, port),
-		},
-		http: http.DefaultClient,
+	if portNum, err := strconv.ParseInt(port, 10, 32); err != nil || portNum > 65535 || portNum < 0 {
+		return OllamaHost{}, ErrInvalidHostPort
+	}
+
+	return OllamaHost{
+		Scheme: scheme,
+		Host:   host,
+		Port:   port,
 	}, nil
 }
 

api/client_test.go

@@ -1,6 +1,12 @@
 package api
 
-import "testing"
+import (
+	"fmt"
+	"net"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
 
 func TestClientFromEnvironment(t *testing.T) {
 	type testCase struct {
@@ -40,4 +46,40 @@ func TestClientFromEnvironment(t *testing.T) {
 			}
 		})
 	}
+
+	hostTestCases := map[string]*testCase{
+		"empty":               {value: "", expect: "127.0.0.1:11434"},
+		"only address":        {value: "1.2.3.4", expect: "1.2.3.4:11434"},
+		"only port":           {value: ":1234", expect: ":1234"},
+		"address and port":    {value: "1.2.3.4:1234", expect: "1.2.3.4:1234"},
+		"hostname":            {value: "example.com", expect: "example.com:11434"},
+		"hostname and port":   {value: "example.com:1234", expect: "example.com:1234"},
+		"zero port":           {value: ":0", expect: ":0"},
+		"too large port":      {value: ":66000", err: ErrInvalidHostPort},
+		"too small port":      {value: ":-1", err: ErrInvalidHostPort},
+		"ipv6 localhost":      {value: "[::1]", expect: "[::1]:11434"},
+		"ipv6 world open":     {value: "[::]", expect: "[::]:11434"},
+		"ipv6 no brackets":    {value: "::1", expect: "[::1]:11434"},
+		"ipv6 + port":         {value: "[::1]:1337", expect: "[::1]:1337"},
+		"extra space":         {value: " 1.2.3.4 ", expect: "1.2.3.4:11434"},
+		"extra quotes":        {value: "\"1.2.3.4\"", expect: "1.2.3.4:11434"},
+		"extra space+quotes":  {value: " \" 1.2.3.4 \" ", expect: "1.2.3.4:11434"},
+		"extra single quotes": {value: "'1.2.3.4'", expect: "1.2.3.4:11434"},
+	}
+
+	for k, v := range hostTestCases {
+		t.Run(k, func(t *testing.T) {
+			t.Setenv("OLLAMA_HOST", v.value)
+
+			oh, err := GetOllamaHost()
+			if err != v.err {
+				t.Fatalf("expected %s, got %s", v.err, err)
+			}
+
+			if err == nil {
+				host := net.JoinHostPort(oh.Host, oh.Port)
+				assert.Equal(t, v.expect, host, fmt.Sprintf("%s: expected %s, got %s", k, v.expect, host))
+			}
+		})
+	}
 }

api/types.go

@@ -309,6 +309,7 @@ func (m *Metrics) Summary() {
 }
 
 var ErrInvalidOpts = errors.New("invalid options")
+var ErrInvalidHostPort = errors.New("invalid port specified in OLLAMA_HOST")
 
 func (opts *Options) FromMap(m map[string]interface{}) error {
 	valueOpts := reflect.ValueOf(opts).Elem() // names of the fields in the options struct

app/.gitignore

@@ -1,2 +1 @@
 ollama.syso
-app

app/AppDelegate.h

@@ -1,7 +0,0 @@
-#import <Cocoa/Cocoa.h>
-
-@interface AppDelegate : NSObject <NSApplicationDelegate>
-
-- (void)applicationDidFinishLaunching:(NSNotification *)aNotification;
-
-@end

app/README.md

@@ -1,6 +1,10 @@
 # Ollama App
 
-## macOS
+## Linux
+
+TODO
+
+## MacOS
 
 TODO
 

app/app_darwin.go

@@ -1,76 +0,0 @@
-package main
-
-// #cgo CFLAGS: -x objective-c
-// #cgo LDFLAGS: -framework Cocoa -framework LocalAuthentication -framework ServiceManagement
-// #include "app_darwin.h"
-import "C"
-import (
-	"context"
-	"fmt"
-	"log/slog"
-	"os"
-	"path/filepath"
-	"syscall"
-)
-
-func init() {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		panic(err)
-	}
-
-	ServerLogFile = filepath.Join(home, ".ollama", "logs", "server.log")
-}
-
-func run() {
-	initLogging()
-	slog.Info("ollama macOS app started")
-
-	// Ask to move to applications directory
-	moving := C.askToMoveToApplications()
-	if moving {
-		return
-	}
-
-	C.killOtherInstances()
-
-	code := C.installSymlink()
-	if code != 0 {
-		slog.Error("Failed to install symlink")
-	}
-
-	exe, err := os.Executable()
-	if err != nil {
-		panic(err)
-	}
-
-	var options ServerOptions
-
-	ctx, cancel := context.WithCancel(context.Background())
-	var done chan int
-
-	done, err = SpawnServer(ctx, filepath.Join(filepath.Dir(exe), "..", "Resources", "ollama"), options)
-	if err != nil {
-		slog.Error(fmt.Sprintf("Failed to spawn ollama server %s", err))
-		done = make(chan int, 1)
-		done <- 1
-	}
-
-	// Run the native macOS app
-	// Note: this will block until the app is closed
-	C.run()
-
-	slog.Info("ollama macOS app closed")
-
-	cancel()
-	slog.Info("Waiting for ollama server to shutdown...")
-	if done != nil {
-		<-done
-	}
-	slog.Info("Ollama app exiting")
-}
-
-//export Quit
-func Quit() {
-	syscall.Kill(os.Getpid(), syscall.SIGTERM)
-}

app/app_darwin.h

@@ -1,13 +0,0 @@
-#import <Cocoa/Cocoa.h>
-
-@interface AppDelegate : NSObject <NSApplicationDelegate>
-- (void)applicationDidFinishLaunching:(NSNotification *)aNotification;
-@end
-
-void run();
-void killOtherInstances();
-bool askToMoveToApplications();
-int createSymlinkWithAuthorization();
-int installSymlink();
-extern void Restart();
-extern void Quit();

app/app_darwin.m

@@ -1,282 +0,0 @@
-#import <AppKit/AppKit.h>
-#import <Cocoa/Cocoa.h>
-#import <CoreServices/CoreServices.h>
-#import <Security/Security.h>
-#import <ServiceManagement/ServiceManagement.h>
-#import "app_darwin.h"
-
-@interface AppDelegate ()
-
-@property (strong, nonatomic) NSStatusItem *statusItem;
-
-@end
-
-@implementation AppDelegate
-
-- (void)applicationDidFinishLaunching:(NSNotification *)aNotification {
-    // show status menu
-    NSMenu *menu = [[NSMenu alloc] init];
-
-    NSMenuItem *aboutMenuItem = [[NSMenuItem alloc] initWithTitle:@"About Ollama" action:@selector(aboutOllama) keyEquivalent:@""];
-    [aboutMenuItem setTarget:self];
-    [menu addItem:aboutMenuItem];
-
-    // Settings submenu
-    NSMenu *settingsMenu = [[NSMenu alloc] initWithTitle:@"Settings"];
-
-    // Submenu items
-    NSMenuItem *chooseModelDirectoryItem = [[NSMenuItem alloc] initWithTitle:@"Choose model directory..." action:@selector(chooseModelDirectory) keyEquivalent:@""];
-    [chooseModelDirectoryItem setTarget:self];
-    [chooseModelDirectoryItem setEnabled:YES];
-    [settingsMenu addItem:chooseModelDirectoryItem];
-
-    NSMenuItem *exposeExternallyItem = [[NSMenuItem alloc] initWithTitle:@"Allow external connections" action:@selector(toggleExposeExternally:) keyEquivalent:@""];
-    [exposeExternallyItem setTarget:self];
-    [exposeExternallyItem setState:NSOffState]; // Set initial state to off
-    [exposeExternallyItem setEnabled:YES];
-    [settingsMenu addItem:exposeExternallyItem];
-
-    NSMenuItem *allowCrossOriginItem = [[NSMenuItem alloc] initWithTitle:@"Allow browser requests" action:@selector(toggleCrossOrigin:) keyEquivalent:@""];
-    [allowCrossOriginItem setTarget:self];
-    [allowCrossOriginItem setState:NSOffState]; // Set initial state to off
-    [allowCrossOriginItem setEnabled:YES];
-    [settingsMenu addItem:allowCrossOriginItem];
-
-    NSMenuItem *settingsMenuItem = [[NSMenuItem alloc] initWithTitle:@"Settings" action:nil keyEquivalent:@""];
-    [settingsMenuItem setSubmenu:settingsMenu];
-    [menu addItem:settingsMenuItem];
-
-    [menu addItemWithTitle:@"Quit Ollama" action:@selector(quit) keyEquivalent:@"q"];
-
-    self.statusItem = [[NSStatusBar systemStatusBar] statusItemWithLength:NSVariableStatusItemLength];
-    [self.statusItem addObserver:self forKeyPath:@"button.effectiveAppearance" options:NSKeyValueObservingOptionNew|NSKeyValueObservingOptionInitial context:nil];
-
-    self.statusItem.menu = menu;
-    [self showIcon];
-}
-
-- (void)aboutOllama {
-    [[NSApplication sharedApplication] orderFrontStandardAboutPanel:nil];
-}
-
-- (void)toggleCrossOrigin:(id)sender {
-    NSMenuItem *item = (NSMenuItem *)sender;
-    if ([item state] == NSOffState) {
-        // Do something when cross-origin requests are allowed
-        [item setState:NSOnState];
-    } else {
-        // Do something when cross-origin requests are disallowed
-        [item setState:NSOffState];
-    }
-}
-
-- (void)toggleExposeExternally:(id)sender {
-    NSMenuItem *item = (NSMenuItem *)sender;
-    if ([item state] == NSOffState) {
-        // Do something when Ollama is exposed externally
-        [item setState:NSOnState];
-    } else {
-        // Do something when Ollama is not exposed externally
-        [item setState:NSOffState];
-    }
-}
-
-- (void)chooseModelDirectory {
-    NSOpenPanel *openPanel = [NSOpenPanel openPanel];
-    [openPanel setCanChooseFiles:NO];
-    [openPanel setCanChooseDirectories:YES];
-    [openPanel setAllowsMultipleSelection:NO];
-
-    NSInteger result = [openPanel runModal];
-    if (result == NSModalResponseOK) {
-        NSURL *selectedDirectoryURL = [openPanel URLs].firstObject;
-        // Do something with the selected directory URL
-    }
-}
-
--(void) showIcon {
-    NSAppearance* appearance = self.statusItem.button.effectiveAppearance;
-    NSString* appearanceName = (NSString*)(appearance.name);
-    NSString* iconName = [[appearanceName lowercaseString] containsString:@"dark"] ? @"iconDark" : @"icon";
-    NSImage* statusImage = [NSImage imageNamed:iconName];
-    [statusImage setTemplate:YES];
-    self.statusItem.button.image = statusImage;
-}
-
--(void)observeValueForKeyPath:(NSString *)keyPath ofObject:(id)object change:(NSDictionary<NSKeyValueChangeKey,id> *)change context:(void *)context {
-    [self showIcon];
-}
-
-- (void)quit {
-    [NSApp stop:nil];
-}
-
-@end
-
-void run() {
-    @autoreleasepool {
-        [NSApplication sharedApplication];
-        AppDelegate *appDelegate = [[AppDelegate alloc] init];
-        [NSApp setDelegate:appDelegate];
-        [NSApp run];
-    }
-}
-
-// killOtherInstances kills all other instances of the app currently
-// running. This way we can ensure that only the most recently started
-// instance of Ollama is running
-void killOtherInstances() {
-    pid_t pid = getpid();
-    NSArray *all = [[NSWorkspace sharedWorkspace] runningApplications];
-    NSMutableArray *apps = [NSMutableArray array];
-
-    for (NSRunningApplication *app in all) {
-        if ([app.bundleIdentifier isEqualToString:[[NSBundle mainBundle] bundleIdentifier]] ||
-            [app.bundleIdentifier isEqualToString:@"ai.ollama.ollama"] ||
-            [app.bundleIdentifier isEqualToString:@"com.electron.ollama"]) {
-            if (app.processIdentifier != pid) {
-                [apps addObject:app];
-            }
-        }
-    }
-
-    for (NSRunningApplication *app in apps) {
-        kill(app.processIdentifier, SIGTERM);
-    }
-
-    NSDate *startTime = [NSDate date];
-    for (NSRunningApplication *app in apps) {
-        while (!app.terminated) {
-            if (-[startTime timeIntervalSinceNow] >= 5) {
-                kill(app.processIdentifier, SIGKILL);
-                break;
-            }
-
-            [[NSRunLoop currentRunLoop] runUntilDate:[NSDate dateWithTimeIntervalSinceNow:0.1]];
-        }
-    }
-}
-
-bool askToMoveToApplications() {
-    NSString *bundlePath = [[NSBundle mainBundle] bundlePath];
-    if ([bundlePath hasPrefix:@"/Applications"]) {
-        return false;
-    }
-
-    NSAlert *alert = [[NSAlert alloc] init];
-    [alert setMessageText:@"Move to Applications?"];
-    [alert setInformativeText:@"Ollama works best when run from the Applications directory."];
-    [alert addButtonWithTitle:@"Move to Applications"];
-    [alert addButtonWithTitle:@"Don't move"];
-
-    [NSApp activateIgnoringOtherApps:YES];
-
-    if ([alert runModal] != NSAlertFirstButtonReturn) {
-        return false;
-    }
-
-    // move to applications
-    NSString *applicationsPath = @"/Applications";
-    NSString *newPath = [applicationsPath stringByAppendingPathComponent:@"Ollama.app"];
-    NSFileManager *fileManager = [NSFileManager defaultManager];
-
-    // Check if the newPath already exists
-    if ([fileManager fileExistsAtPath:newPath]) {
-        NSError *removeError = nil;
-        [fileManager removeItemAtPath:newPath error:&removeError];
-        if (removeError) {
-            NSLog(@"Error removing file at %@: %@", newPath, removeError);
-            return false; // or handle the error
-        }
-    }
-
-    NSError *moveError = nil;
-    [fileManager moveItemAtPath:bundlePath toPath:newPath error:&moveError];
-    if (moveError) {
-        NSLog(@"Error moving file from %@ to %@: %@", bundlePath, newPath, moveError);
-        return false;
-    }
-
-    NSLog(@"Opening %@", newPath);
-    NSError *error = nil;
-    NSWorkspace *workspace = [NSWorkspace sharedWorkspace];
-#pragma clang diagnostic ignored "-Wdeprecated-declarations"
-    [workspace launchApplicationAtURL:[NSURL fileURLWithPath:newPath]
-               options:NSWorkspaceLaunchNewInstance | NSWorkspaceLaunchDefault
-               configuration:@{}
-               error:&error];
-
-    return true;
-}
-
-int installSymlink() {
-    NSString *linkPath = @"/usr/local/bin/ollama";
-    NSError *error = nil;
-
-    NSFileManager *fileManager = [NSFileManager defaultManager];
-    NSString *symlinkPath = [fileManager destinationOfSymbolicLinkAtPath:linkPath error:&error];
-    NSString *bundlePath = [[NSBundle mainBundle] bundlePath];
-    NSString *execPath = [[NSBundle mainBundle] executablePath];
-    NSString *resPath = [[NSBundle mainBundle] pathForResource:@"ollama" ofType:nil];
-
-    // if the symlink already exists and points to the right place, don't prompt
-    if ([symlinkPath isEqualToString:resPath]) {
-        NSLog(@"symbolic link already exists and points to the right place");
-        return 0;
-    }
-
-    NSString *authorizationPrompt = @"Ollama is trying to install its command line interface (CLI) tool.";
-
-    AuthorizationRef auth = NULL;
-    OSStatus createStatus = AuthorizationCreate(NULL, kAuthorizationEmptyEnvironment, kAuthorizationFlagDefaults, &auth);
-    if (createStatus != errAuthorizationSuccess) {
-        NSLog(@"Error creating authorization");
-        return -1;
-    }
-
-    NSString * bundleIdentifier = [[NSBundle mainBundle] bundleIdentifier];
-    NSString *rightNameString = [NSString stringWithFormat:@"%@.%@", bundleIdentifier, @"auth3"];
-    const char *rightName = rightNameString.UTF8String;
-
-    OSStatus getRightResult = AuthorizationRightGet(rightName, NULL);
-    if (getRightResult == errAuthorizationDenied) {
-        if (AuthorizationRightSet(auth, rightName, (__bridge CFTypeRef _Nonnull)(@(kAuthorizationRuleAuthenticateAsAdmin)), (__bridge CFStringRef _Nullable)(authorizationPrompt), NULL, NULL) != errAuthorizationSuccess) {
-            NSLog(@"Failed to set right");
-            return -1;
-        }
-    }
-
-    AuthorizationItem right = { .name = rightName, .valueLength = 0, .value = NULL, .flags = 0 };
-    AuthorizationRights rights = { .count = 1, .items = &right };
-    AuthorizationFlags flags = (AuthorizationFlags)(kAuthorizationFlagExtendRights | kAuthorizationFlagInteractionAllowed);
-    AuthorizationItem iconAuthorizationItem = {.name = kAuthorizationEnvironmentIcon, .valueLength = 0, .value = NULL, .flags = 0};
-    AuthorizationEnvironment authorizationEnvironment = {.count = 0, .items = NULL};
-
-    BOOL failedToUseSystemDomain = NO;
-    OSStatus copyStatus = AuthorizationCopyRights(auth, &rights, &authorizationEnvironment, flags, NULL);
-    if (copyStatus != errAuthorizationSuccess) {
-        failedToUseSystemDomain = YES;
-
-        if (copyStatus == errAuthorizationCanceled) {
-            NSLog(@"User cancelled authorization");
-            return -1;
-        } else {
-            NSLog(@"Failed copying system domain rights: %d", copyStatus);
-            return -1;
-        }
-    }
-
-    const char *toolPath = "/bin/ln";
-    const char *args[] = {"-s", "-F", [resPath UTF8String], "/usr/local/bin/ollama", NULL};
-    FILE *pipe = NULL;
-
-#pragma clang diagnostic ignored "-Wdeprecated-declarations"
-    OSStatus status = AuthorizationExecuteWithPrivileges(auth, toolPath, kAuthorizationFlagDefaults, (char *const *)args, &pipe);
-    if (status != errAuthorizationSuccess) {
-        NSLog(@"Failed to create symlink");
-        return -1;
-    }
-
-    AuthorizationFree(auth, kAuthorizationFlagDestroyRights);
-    return 0;
-}

app/app_windows.go

@@ -1,166 +0,0 @@
-package main
-
-import (
-	"context"
-	"errors"
-	"fmt"
-	"log"
-	"log/slog"
-	"os"
-	"os/exec"
-	"os/signal"
-	"path/filepath"
-	"strings"
-	"syscall"
-
-	"github.com/ollama/ollama/app/lifecycle"
-	"github.com/ollama/ollama/app/store"
-	"github.com/ollama/ollama/app/tray"
-	"github.com/ollama/ollama/app/updater"
-)
-
-func init() {
-	AppName += ".exe"
-	CLIName += ".exe"
-	// Logs, configs, downloads go to LOCALAPPDATA
-	localAppData := os.Getenv("LOCALAPPDATA")
-	AppDataDir = filepath.Join(localAppData, "Ollama")
-	AppLogFile = filepath.Join(AppDataDir, "app.log")
-	ServerLogFile = filepath.Join(AppDataDir, "server.log")
-
-	// Executables are stored in APPDATA
-	AppDir = filepath.Join(localAppData, "Programs", "Ollama")
-
-	// Make sure we have PATH set correctly for any spawned children
-	paths := strings.Split(os.Getenv("PATH"), ";")
-	// Start with whatever we find in the PATH/LD_LIBRARY_PATH
-	found := false
-	for _, path := range paths {
-		d, err := filepath.Abs(path)
-		if err != nil {
-			continue
-		}
-		if strings.EqualFold(AppDir, d) {
-			found = true
-		}
-	}
-	if !found {
-		paths = append(paths, AppDir)
-
-		pathVal := strings.Join(paths, ";")
-		slog.Debug("setting PATH=" + pathVal)
-		err := os.Setenv("PATH", pathVal)
-		if err != nil {
-			slog.Error(fmt.Sprintf("failed to update PATH: %s", err))
-		}
-	}
-
-	// Make sure our logging dir exists
-	_, err := os.Stat(AppDataDir)
-	if errors.Is(err, os.ErrNotExist) {
-		if err := os.MkdirAll(AppDataDir, 0o755); err != nil {
-			slog.Error(fmt.Sprintf("create ollama dir %s: %v", AppDataDir, err))
-		}
-	}
-}
-
-func ShowLogs() {
-	cmd_path := "c:\\Windows\\system32\\cmd.exe"
-	slog.Debug(fmt.Sprintf("viewing logs with start %s", AppDataDir))
-	cmd := exec.Command(cmd_path, "/c", "start", AppDataDir)
-	cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: false, CreationFlags: 0x08000000}
-	err := cmd.Start()
-	if err != nil {
-		slog.Error(fmt.Sprintf("Failed to open log dir: %s", err))
-	}
-}
-
-func Start() {
-	cmd_path := "c:\\Windows\\system32\\cmd.exe"
-	slog.Debug(fmt.Sprintf("viewing logs with start %s", AppDataDir))
-	cmd := exec.Command(cmd_path, "/c", "start", AppDataDir)
-	cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: false, CreationFlags: 0x08000000}
-	err := cmd.Start()
-	if err != nil {
-		slog.Error(fmt.Sprintf("Failed to open log dir: %s", err))
-	}
-}
-
-func run() {
-	initLogging()
-
-	slog.Info("ollama windows app started")
-
-	ctx, cancel := context.WithCancel(context.Background())
-	var done chan int
-
-	t, err := tray.NewTray()
-	if err != nil {
-		log.Fatalf("Failed to start: %s", err)
-	}
-	callbacks := t.GetCallbacks()
-
-	signals := make(chan os.Signal, 1)
-	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
-
-	go func() {
-		slog.Debug("starting callback loop")
-		for {
-			select {
-			case <-callbacks.Quit:
-				slog.Debug("quit called")
-				t.Quit()
-			case <-signals:
-				slog.Debug("shutting down due to signal")
-				t.Quit()
-			case <-callbacks.Update:
-				err := updater.DoUpgrade(cancel, done)
-				if err != nil {
-					slog.Warn(fmt.Sprintf("upgrade attempt failed: %s", err))
-				}
-			case <-callbacks.ShowLogs:
-				ShowLogs()
-			case <-callbacks.DoFirstUse:
-				err := lifecycle.GetStarted()
-				if err != nil {
-					slog.Warn(fmt.Sprintf("Failed to launch getting started shell: %s", err))
-				}
-			}
-		}
-	}()
-
-	if !store.GetFirstTimeRun() {
-		slog.Debug("First time run")
-		err = t.DisplayFirstUseNotification()
-		if err != nil {
-			slog.Debug(fmt.Sprintf("XXX failed to display first use notification %v", err))
-		}
-		store.SetFirstTimeRun(true)
-	} else {
-		slog.Debug("Not first time, skipping first run notification")
-	}
-
-	if isServerRunning(ctx) {
-		slog.Info("Detected another instance of ollama running, exiting")
-		os.Exit(1)
-	}
-
-	done, err = SpawnServer(ctx, CLIName)
-	if err != nil {
-		// TODO - should we retry in a backoff loop?
-		// TODO - should we pop up a warning and maybe add a menu item to view application logs?
-		slog.Error(fmt.Sprintf("Failed to spawn ollama server %s", err))
-		done = make(chan int, 1)
-		done <- 1
-	}
-
-	updater.StartBackgroundUpdaterChecker(ctx, t.UpdateAvailable)
-
-	t.Run()
-	cancel()
-	slog.Info("Waiting for ollama server to shutdown...")
-	if done != nil {
-		<-done
-	}
-	slog.Info("Ollama app exiting")
-}

app/darwin/Ollama.app/Contents/Info.plist

@@ -1,40 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
-<plist version="1.0">
-  <dict>
-    <key>CFBundleDisplayName</key>
-    <string>Ollama</string>
-    <key>CFBundleExecutable</key>
-    <string>Ollama</string>
-    <key>CFBundleIconFile</key>
-    <string>icon.icns</string>
-    <key>CFBundleIdentifier</key>
-    <string>com.ollama.ollama</string>
-    <key>CFBundleInfoDictionaryVersion</key>
-    <string>6.0</string>
-    <key>CFBundleName</key>
-    <string>Ollama</string>
-    <key>CFBundlePackageType</key>
-    <string>APPL</string>
-    <key>CFBundleShortVersionString</key>
-    <string>0.0.0</string>
-    <key>CFBundleVersion</key>
-    <string>0.0.0</string>
-    <key>DTCompiler</key>
-    <string>com.apple.compilers.llvm.clang.1_0</string>
-    <key>DTSDKBuild</key>
-    <string>22E245</string>
-    <key>DTSDKName</key>
-    <string>macosx13.3</string>
-    <key>DTXcode</key>
-    <string>1431</string>
-    <key>DTXcodeBuild</key>
-    <string>14E300c</string>
-    <key>LSApplicationCategoryType</key>
-    <string>public.app-category.developer-tools</string>
-    <key>LSMinimumSystemVersion</key>
-    <string>11.0</string>
-    <key>LSUIElement</key>
-    <true/>
-  </dict>
-</plist>

app/lifecycle/getstarted_nonwindows.go

@@ -1,3 +1,5 @@
+//go:build !windows
+
 package lifecycle
 
 import "fmt"

app/lifecycle/lifecycle.go

@@ -0,0 +1,92 @@
+package lifecycle
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"log/slog"
+	"os"
+	"os/signal"
+	"syscall"
+
+	"github.com/ollama/ollama/app/store"
+	"github.com/ollama/ollama/app/tray"
+)
+
+func Run() {
+	InitLogging()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	var done chan int
+
+	t, err := tray.NewTray()
+	if err != nil {
+		log.Fatalf("Failed to start: %s", err)
+	}
+	callbacks := t.GetCallbacks()
+
+	signals := make(chan os.Signal, 1)
+	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
+
+	go func() {
+		slog.Debug("starting callback loop")
+		for {
+			select {
+			case <-callbacks.Quit:
+				slog.Debug("quit called")
+				t.Quit()
+			case <-signals:
+				slog.Debug("shutting down due to signal")
+				t.Quit()
+			case <-callbacks.Update:
+				err := DoUpgrade(cancel, done)
+				if err != nil {
+					slog.Warn(fmt.Sprintf("upgrade attempt failed: %s", err))
+				}
+			case <-callbacks.ShowLogs:
+				ShowLogs()
+			case <-callbacks.DoFirstUse:
+				err := GetStarted()
+				if err != nil {
+					slog.Warn(fmt.Sprintf("Failed to launch getting started shell: %s", err))
+				}
+			}
+		}
+	}()
+
+	// Are we first use?
+	if !store.GetFirstTimeRun() {
+		slog.Debug("First time run")
+		err = t.DisplayFirstUseNotification()
+		if err != nil {
+			slog.Debug(fmt.Sprintf("XXX failed to display first use notification %v", err))
+		}
+		store.SetFirstTimeRun(true)
+	} else {
+		slog.Debug("Not first time, skipping first run notification")
+	}
+
+	if IsServerRunning(ctx) {
+		slog.Info("Detected another instance of ollama running, exiting")
+		os.Exit(1)
+	} else {
+		done, err = SpawnServer(ctx, CLIName)
+		if err != nil {
+			// TODO - should we retry in a backoff loop?
+			// TODO - should we pop up a warning and maybe add a menu item to view application logs?
+			slog.Error(fmt.Sprintf("Failed to spawn ollama server %s", err))
+			done = make(chan int, 1)
+			done <- 1
+		}
+	}
+
+	StartBackgroundUpdaterChecker(ctx, t.UpdateAvailable)
+
+	t.Run()
+	cancel()
+	slog.Info("Waiting for ollama server to shutdown...")
+	if done != nil {
+		<-done
+	}
+	slog.Info("Ollama app exiting")
+}

app/lifecycle/logging.go

@@ -1,4 +1,4 @@
-package main
+package lifecycle
 
 import (
 	"fmt"
@@ -7,7 +7,7 @@ import (
 	"path/filepath"
 )
 
-func initLogging() {
+func InitLogging() {
 	level := slog.LevelInfo
 
 	if debug := os.Getenv("OLLAMA_DEBUG"); debug != "" {
@@ -41,4 +41,6 @@ func initLogging() {
 	})
 
 	slog.SetDefault(slog.New(handler))
+
+	slog.Info("ollama app started")
 }

app/lifecycle/logging_nonwindows.go

@@ -0,0 +1,9 @@
+//go:build !windows
+
+package lifecycle
+
+import "log/slog"
+
+func ShowLogs() {
+	slog.Warn("ShowLogs not yet implemented")
+}

app/lifecycle/logging_windows.go

@@ -0,0 +1,19 @@
+package lifecycle
+
+import (
+	"fmt"
+	"log/slog"
+	"os/exec"
+	"syscall"
+)
+
+func ShowLogs() {
+	cmd_path := "c:\\Windows\\system32\\cmd.exe"
+	slog.Debug(fmt.Sprintf("viewing logs with start %s", AppDataDir))
+	cmd := exec.Command(cmd_path, "/c", "start", AppDataDir)
+	cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: false, CreationFlags: 0x08000000}
+	err := cmd.Start()
+	if err != nil {
+		slog.Error(fmt.Sprintf("Failed to open log dir: %s", err))
+	}
+}

app/lifecycle/paths.go

@@ -70,5 +70,10 @@ func init() {
 			}
 		}
 
+	} else if runtime.GOOS == "darwin" {
+		// TODO
+		AppName += ".app"
+		// } else if runtime.GOOS == "linux" {
+		// TODO
 	}
 }

app/lifecycle/server.go

@@ -1,4 +1,4 @@
-package main
+package lifecycle
 
 import (
 	"context"
@@ -14,28 +14,37 @@ import (
 	"github.com/ollama/ollama/api"
 )
 
-type ServerOptions struct {
-	Cors       bool
-	Expose     bool
-	ModelsPath string
+func getCLIFullPath(command string) string {
+	cmdPath := ""
+	appExe, err := os.Executable()
+	if err == nil {
+		cmdPath = filepath.Join(filepath.Dir(appExe), command)
+		_, err := os.Stat(cmdPath)
+		if err == nil {
+			return cmdPath
+		}
+	}
+	cmdPath, err = exec.LookPath(command)
+	if err == nil {
+		_, err := os.Stat(cmdPath)
+		if err == nil {
+			return cmdPath
+		}
+	}
+	pwd, err := os.Getwd()
+	if err == nil {
+		cmdPath = filepath.Join(pwd, command)
+		_, err = os.Stat(cmdPath)
+		if err == nil {
+			return cmdPath
+		}
+	}
+
+	return command
 }
 
-func start(ctx context.Context, command string, options ServerOptions) (*exec.Cmd, error) {
-	cmd := getCmd(ctx, command)
-
-	// set environment variables
-	if options.ModelsPath != "" {
-		cmd.Env = append(cmd.Env, fmt.Sprintf("OLLAMA_MODELS=%s", options.ModelsPath))
-	}
-
-	if options.Cors {
-		cmd.Env = append(cmd.Env, "OLLAMA_ORIGINS=*")
-	}
-
-	if options.Expose {
-		cmd.Env = append(cmd.Env, "OLLAMA_HOST=0.0.0.0")
-	}
-
+func start(ctx context.Context, command string) (*exec.Cmd, error) {
+	cmd := getCmd(ctx, getCLIFullPath(command))
 	stdout, err := cmd.StdoutPipe()
 	if err != nil {
 		return nil, fmt.Errorf("failed to spawn server stdout pipe: %w", err)
@@ -50,6 +59,20 @@ func start(ctx context.Context, command string, options ServerOptions) (*exec.Cm
 	if err != nil {
 		return nil, fmt.Errorf("failed to create server log: %w", err)
 	}
+
+	logDir := filepath.Dir(ServerLogFile)
+	_, err = os.Stat(logDir)
+	if err != nil {
+		if !errors.Is(err, os.ErrNotExist) {
+			return nil, fmt.Errorf("stat ollama server log dir %s: %v", logDir, err)
+
+		}
+
+		if err := os.MkdirAll(logDir, 0o755); err != nil {
+			return nil, fmt.Errorf("create ollama server log dir %s: %v", logDir, err)
+		}
+	}
+
 	go func() {
 		defer logFile.Close()
 		io.Copy(logFile, stdout) //nolint:errcheck
@@ -103,25 +126,20 @@ func start(ctx context.Context, command string, options ServerOptions) (*exec.Cm
 	return cmd, nil
 }
 
-func SpawnServer(ctx context.Context, command string, options ServerOptions) (chan int, error) {
-	logDir := filepath.Dir(ServerLogFile)
-	_, err := os.Stat(logDir)
-	if errors.Is(err, os.ErrNotExist) {
-		if err := os.MkdirAll(logDir, 0o755); err != nil {
-			return nil, fmt.Errorf("create ollama server log dir %s: %v", logDir, err)
-		}
-	}
-
+func SpawnServer(ctx context.Context, command string) (chan int, error) {
 	done := make(chan int)
 
 	go func() {
 		// Keep the server running unless we're shuttind down the app
 		crashCount := 0
 		for {
-			slog.Info(fmt.Sprintf("starting server..."))
-			cmd, err := start(ctx, command, options)
+			slog.Info("starting server...")
+			cmd, err := start(ctx, command)
 			if err != nil {
+				crashCount++
 				slog.Error(fmt.Sprintf("failed to start server %s", err))
+				time.Sleep(500 * time.Millisecond * time.Duration(crashCount))
+				continue
 			}
 
 			cmd.Wait() //nolint:errcheck
@@ -147,7 +165,7 @@ func SpawnServer(ctx context.Context, command string, options ServerOptions) (ch
 	return done, nil
 }
 
-func isServerRunning(ctx context.Context) bool {
+func IsServerRunning(ctx context.Context) bool {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		slog.Info("unable to connect to server")

app/lifecycle/server_unix.go

@@ -1,4 +1,6 @@
-package main
+//go:build !windows
+
+package lifecycle
 
 import (
 	"context"

app/lifecycle/server_windows.go

@@ -1,4 +1,4 @@
-package main
+package lifecycle
 
 import (
 	"context"

app/lifecycle/updater.go

@@ -1,4 +1,4 @@
-package updater
+package lifecycle
 
 import (
 	"context"
@@ -22,10 +22,6 @@ import (
 	"github.com/ollama/ollama/version"
 )
 
-var (
-	UpdateStageDir string
-)
-
 var (
 	UpdateCheckURLBase  = "https://ollama.com/api/update"
 	UpdateDownloaded    = false
@@ -127,7 +123,7 @@ func DownloadNewRelease(ctx context.Context, updateResp UpdateResponse) error {
 		slog.Debug("no etag detected, falling back to filename based dedup")
 		etag = "_"
 	}
-	filename := "OllamaSetup.exe"
+	filename := Installer
 	_, params, err := mime.ParseMediaType(resp.Header.Get("content-disposition"))
 	if err == nil {
 		filename = params["filename"]

app/lifecycle/updater_nonwindows.go

@@ -1,4 +1,6 @@
-package updater
+//go:build !windows
+
+package lifecycle
 
 import (
 	"context"

app/lifecycle/updater_windows.go

@@ -1,4 +1,4 @@
-package updater
+package lifecycle
 
 import (
 	"context"
@@ -9,13 +9,7 @@ import (
 	"path/filepath"
 )
 
-func init() {
-	UpdateStageDir = filepath.Join(os.Getenv("LOCALAPPDATA"), "Ollama", "updates")
-}
-
 func DoUpgrade(cancel context.CancelFunc, done chan int) error {
-	logFile := filepath.Join(os.Getenv("LOCALAPPDATA"), "Ollama", "upgrade.log")
-
 	files, err := filepath.Glob(filepath.Join(UpdateStageDir, "*", "*.exe")) // TODO generalize for multiplatform
 	if err != nil {
 		return fmt.Errorf("failed to lookup downloads: %s", err)
@@ -29,13 +23,13 @@ func DoUpgrade(cancel context.CancelFunc, done chan int) error {
 	installerExe := files[0]
 
 	slog.Info("starting upgrade with " + installerExe)
-	slog.Info("upgrade log file " + logFile)
+	slog.Info("upgrade log file " + UpgradeLogFile)
 
 	// When running in debug mode, we'll be "verbose" and let the installer pop up and prompt
 	installArgs := []string{
-		"/CLOSEAPPLICATIONS",             // Quit the tray app if it's still running
-		"/LOG=" + filepath.Base(logFile), // Only relative seems reliable, so set pwd
-		"/FORCECLOSEAPPLICATIONS",        // Force close the tray app - might be needed
+		"/CLOSEAPPLICATIONS",                    // Quit the tray app if it's still running
+		"/LOG=" + filepath.Base(UpgradeLogFile), // Only relative seems reliable, so set pwd
+		"/FORCECLOSEAPPLICATIONS",               // Force close the tray app - might be needed
 	}
 	// When we're not in debug mode, make the upgrade as quiet as possible (no GUI, no prompts)
 	// TODO - temporarily disable since we're pinning in debug mode for the preview
@@ -59,7 +53,7 @@ func DoUpgrade(cancel context.CancelFunc, done chan int) error {
 	}
 
 	slog.Debug(fmt.Sprintf("starting installer: %s %v", installerExe, installArgs))
-	os.Chdir(filepath.Dir(logFile)) //nolint:errcheck
+	os.Chdir(filepath.Dir(UpgradeLogFile)) //nolint:errcheck
 	cmd := exec.Command(installerExe, installArgs...)
 
 	if err := cmd.Start(); err != nil {

app/main.go

@@ -2,15 +2,11 @@ package main
 
 // Compile with the following to get rid of the cmd pop up on windows
 // go build -ldflags="-H windowsgui" .
-var (
-	AppName       string
-	CLIName       string
-	AppDir        string
-	AppDataDir    string
-	AppLogFile    string
-	ServerLogFile string
+
+import (
+	"github.com/ollama/ollama/app/lifecycle"
 )
 
 func main() {
-	run()
+	lifecycle.Run()
 }

app/ollama.iss

@@ -88,8 +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\windows-amd64\*.dll"; DestDir: "{app}"; Flags: ignoreversion 64bit
-Source: "..\dist\windows-amd64\ollama_runners\*"; DestDir: "{app}\ollama_runners"; Flags: ignoreversion 64bit recursesubdirs
+Source: "..\dist\windows-{#ARCH}\*.dll"; DestDir: "{app}"; Flags: ignoreversion 64bit
+Source: "..\dist\windows-{#ARCH}\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
 #if DirExists("..\dist\windows-amd64\rocm")

app/tray/tray_nonwindows.go

@@ -1,3 +1,5 @@
+//go:build !windows
+
 package tray
 
 import (

auth/auth.go

@@ -10,12 +10,44 @@ import (
 	"log/slog"
 	"os"
 	"path/filepath"
+	"strings"
 
 	"golang.org/x/crypto/ssh"
 )
 
 const defaultPrivateKey = "id_ed25519"
 
+func keyPath() (string, error) {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return "", err
+	}
+
+	return filepath.Join(home, ".ollama", defaultPrivateKey), nil
+}
+
+func GetPublicKey() (string, error) {
+	keyPath, err := keyPath()
+	if err != nil {
+		return "", err
+	}
+
+	privateKeyFile, err := os.ReadFile(keyPath)
+	if err != nil {
+		slog.Info(fmt.Sprintf("Failed to load private key: %v", err))
+		return "", err
+	}
+
+	privateKey, err := ssh.ParsePrivateKey(privateKeyFile)
+	if err != nil {
+		return "", err
+	}
+
+	publicKey := ssh.MarshalAuthorizedKey(privateKey.PublicKey())
+
+	return strings.TrimSpace(string(publicKey)), nil
+}
+
 func NewNonce(r io.Reader, length int) (string, error) {
 	nonce := make([]byte, length)
 	if _, err := io.ReadFull(r, nonce); err != nil {
@@ -26,13 +58,11 @@ func NewNonce(r io.Reader, length int) (string, error) {
 }
 
 func Sign(ctx context.Context, bts []byte) (string, error) {
-	home, err := os.UserHomeDir()
+	keyPath, err := keyPath()
 	if err != nil {
 		return "", err
 	}
 
-	keyPath := filepath.Join(home, ".ollama", defaultPrivateKey)
-
 	privateKeyFile, err := os.ReadFile(keyPath)
 	if err != nil {
 		slog.Info(fmt.Sprintf("Failed to load private key: %v", err))

cmd/cmd.go

@@ -32,10 +32,13 @@ import (
 	"golang.org/x/term"
 
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/auth"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/parser"
 	"github.com/ollama/ollama/progress"
 	"github.com/ollama/ollama/server"
+	"github.com/ollama/ollama/types/errtypes"
+	"github.com/ollama/ollama/types/model"
 	"github.com/ollama/ollama/version"
 )
 
@@ -54,12 +57,13 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	p := progress.NewProgress(os.Stderr)
 	defer p.Stop()
 
-	modelfile, err := os.ReadFile(filename)
+	modelfile, err := os.Open(filename)
 	if err != nil {
 		return err
 	}
+	defer modelfile.Close()
 
-	commands, err := parser.Parse(bytes.NewReader(modelfile))
+	commands, err := parser.Parse(modelfile)
 	if err != nil {
 		return err
 	}
@@ -73,10 +77,10 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	spinner := progress.NewSpinner(status)
 	p.Add(status, spinner)
 
-	for _, c := range commands {
-		switch c.Name {
+	for i := range commands {
+		switch commands[i].Name {
 		case "model", "adapter":
-			path := c.Args
+			path := commands[i].Args
 			if path == "~" {
 				path = home
 			} else if strings.HasPrefix(path, "~/") {
@@ -88,7 +92,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 			}
 
 			fi, err := os.Stat(path)
-			if errors.Is(err, os.ErrNotExist) && c.Name == "model" {
+			if errors.Is(err, os.ErrNotExist) && commands[i].Name == "model" {
 				continue
 			} else if err != nil {
 				return err
@@ -111,13 +115,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 				return err
 			}
 
-			name := c.Name
-			if c.Name == "model" {
-				name = "from"
-			}
-
-			re := regexp.MustCompile(fmt.Sprintf(`(?im)^(%s)\s+%s\s*$`, name, c.Args))
-			modelfile = re.ReplaceAll(modelfile, []byte("$1 @"+digest))
+			commands[i].Args = "@"+digest
 		}
 	}
 
@@ -147,7 +145,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 
 	quantization, _ := cmd.Flags().GetString("quantization")
 
-	request := api.CreateRequest{Name: args[0], Modelfile: string(modelfile), Quantization: quantization}
+	request := api.CreateRequest{Name: args[0], Modelfile: parser.Format(commands), Quantization: quantization}
 	if err := client.Create(cmd.Context(), &request, fn); err != nil {
 		return err
 	}
@@ -357,6 +355,47 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 	return generateInteractive(cmd, opts)
 }
 
+func errFromUnknownKey(unknownKeyErr error) error {
+	// find SSH public key in the error message
+	sshKeyPattern := `ssh-\w+ [^\s"]+`
+	re := regexp.MustCompile(sshKeyPattern)
+	matches := re.FindStringSubmatch(unknownKeyErr.Error())
+
+	if len(matches) > 0 {
+		serverPubKey := matches[0]
+
+		localPubKey, err := auth.GetPublicKey()
+		if err != nil {
+			return unknownKeyErr
+		}
+
+		if runtime.GOOS == "linux" && serverPubKey != localPubKey {
+			// try the ollama service public key
+			svcPubKey, err := os.ReadFile("/usr/share/ollama/.ollama/id_ed25519.pub")
+			if err != nil {
+				return unknownKeyErr
+			}
+			localPubKey = strings.TrimSpace(string(svcPubKey))
+		}
+
+		// check if the returned public key matches the local public key, this prevents adding a remote key to the user's account
+		if serverPubKey != localPubKey {
+			return unknownKeyErr
+		}
+
+		var msg strings.Builder
+		msg.WriteString(unknownKeyErr.Error())
+		msg.WriteString("\n\nYour ollama key is:\n")
+		msg.WriteString(localPubKey)
+		msg.WriteString("\nAdd your key at:\n")
+		msg.WriteString("https://ollama.com/settings/keys")
+
+		return errors.New(msg.String())
+	}
+
+	return unknownKeyErr
+}
+
 func PushHandler(cmd *cobra.Command, args []string) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
@@ -404,6 +443,20 @@ func PushHandler(cmd *cobra.Command, args []string) error {
 
 	request := api.PushRequest{Name: args[0], Insecure: insecure}
 	if err := client.Push(cmd.Context(), &request, fn); err != nil {
+		if spinner != nil {
+			spinner.Stop()
+		}
+		if strings.Contains(err.Error(), "access denied") {
+			return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
+		}
+		host := model.ParseName(args[0]).Host
+		isOllamaHost := strings.HasSuffix(host, ".ollama.ai") || strings.HasSuffix(host, ".ollama.com")
+		if strings.Contains(err.Error(), errtypes.UnknownOllamaKeyErrMsg) && isOllamaHost {
+			// the user has not added their ollama key to ollama.com
+			// re-throw an error with a more user-friendly message
+			return errFromUnknownKey(err)
+		}
+
 		return err
 	}
 
@@ -831,19 +884,17 @@ func generate(cmd *cobra.Command, opts runOptions) error {
 }
 
 func RunServer(cmd *cobra.Command, _ []string) error {
-	host, port, err := net.SplitHostPort(strings.Trim(os.Getenv("OLLAMA_HOST"), "\"'"))
+	// retrieve the OLLAMA_HOST environment variable
+	ollamaHost, err := api.GetOllamaHost()
 	if err != nil {
-		host, port = "127.0.0.1", "11434"
-		if ip := net.ParseIP(strings.Trim(os.Getenv("OLLAMA_HOST"), "[]")); ip != nil {
-			host = ip.String()
-		}
+		return err
 	}
 
 	if err := initializeKeypair(); err != nil {
 		return err
 	}
 
-	ln, err := net.Listen("tcp", net.JoinHostPort(host, port))
+	ln, err := net.Listen("tcp", net.JoinHostPort(ollamaHost.Host, ollamaHost.Port))
 	if err != nil {
 		return err
 	}
@@ -1069,7 +1120,7 @@ Environment Variables:
 		RunE:    ListHandler,
 	}
 	copyCmd := &cobra.Command{
-		Use:     "cp SOURCE TARGET",
+		Use:     "cp SOURCE DESTINATION",
 		Short:   "Copy a model",
 		Args:    cobra.ExactArgs(2),
 		PreRunE: checkServerHeartbeat,

cmd/interactive.go

@@ -94,6 +94,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		fmt.Fprintln(os.Stderr, "  /show           Show model information")
 		fmt.Fprintln(os.Stderr, "  /load <model>   Load a session or model")
 		fmt.Fprintln(os.Stderr, "  /save <model>   Save your current session")
+		fmt.Fprintln(os.Stderr, "  /clear          Clear session context")
 		fmt.Fprintln(os.Stderr, "  /bye            Exit")
 		fmt.Fprintln(os.Stderr, "  /?, /help       Help for a command")
 		fmt.Fprintln(os.Stderr, "  /? shortcuts    Help for keyboard shortcuts")
@@ -280,6 +281,10 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			}
 			fmt.Printf("Created new model '%s'\n", args[1])
 			continue
+		case strings.HasPrefix(line, "/clear"):
+			opts.Messages = []api.Message{}
+			fmt.Println("Cleared session context")
+			continue
 		case strings.HasPrefix(line, "/set"):
 			args := strings.Fields(line)
 			if len(args) > 1 {

convert/convert.go

@@ -5,6 +5,7 @@ import (
 	"encoding/binary"
 	"encoding/json"
 	"fmt"
+	"io"
 	"log/slog"
 	"os"
 	"path/filepath"
@@ -47,7 +48,7 @@ type ByteOrder interface {
 type ModelArch interface {
 	GetTensors() error
 	LoadVocab() error
-	WriteGGUF() (string, error)
+	WriteGGUF(io.WriteSeeker) error
 }
 
 type ModelFormat interface {

convert/gemma.go

@@ -94,7 +94,7 @@ func (m *GemmaModel) LoadVocab() error {
 	return nil
 }
 
-func (m *GemmaModel) WriteGGUF() (string, error) {
+func (m *GemmaModel) WriteGGUF(ws io.WriteSeeker) error {
 	kv := llm.KV{
 		"general.architecture":                   "gemma",
 		"general.name":                           m.Name,
@@ -122,16 +122,5 @@ func (m *GemmaModel) WriteGGUF() (string, error) {
 		"tokenizer.ggml.add_eos_token":    false,
 	}
 
-	f, err := os.CreateTemp("", "ollama-gguf")
-	if err != nil {
-		return "", err
-	}
-	defer f.Close()
-
-	mod := llm.NewGGUFV3(m.Params.ByteOrder)
-	if err := mod.Encode(f, kv, m.Tensors); err != nil {
-		return "", err
-	}
-
-	return f.Name(), nil
+	return llm.NewGGUFV3(m.Params.ByteOrder).Encode(ws, kv, m.Tensors)
 }

convert/llama.go

@@ -132,7 +132,7 @@ func (m *LlamaModel) LoadVocab() error {
 	return nil
 }
 
-func (m *LlamaModel) WriteGGUF() (string, error) {
+func (m *LlamaModel) WriteGGUF(ws io.WriteSeeker) error {
 	kv := llm.KV{
 		"general.architecture":                   "llama",
 		"general.name":                           m.Name,
@@ -161,16 +161,9 @@ func (m *LlamaModel) WriteGGUF() (string, error) {
 
 	f, err := os.CreateTemp("", "ollama-gguf")
 	if err != nil {
-		return "", err
+		return err
 	}
 	defer f.Close()
 
-	mod := llm.NewGGUFV3(m.Params.ByteOrder)
-	if err := mod.Encode(f, kv, m.Tensors); err != nil {
-		return "", err
-	}
-
-	slog.Debug(fmt.Sprintf("gguf file = %s", f.Name()))
-
-	return f.Name(), nil
+	return llm.NewGGUFV3(m.Params.ByteOrder).Encode(f, kv, m.Tensors)
 }

convert/mistral.go

@@ -132,7 +132,7 @@ func (m *MistralModel) LoadVocab() error {
 	return nil
 }
 
-func (m *MistralModel) WriteGGUF() (string, error) {
+func (m *MistralModel) WriteGGUF(ws io.WriteSeeker) error {
 	kv := llm.KV{
 		"general.architecture":                   "llama",
 		"general.name":                           m.Name,
@@ -158,16 +158,5 @@ func (m *MistralModel) WriteGGUF() (string, error) {
 		"tokenizer.ggml.unknown_token_id": uint32(0),
 	}
 
-	f, err := os.CreateTemp("", "ollama-gguf")
-	if err != nil {
-		return "", err
-	}
-	defer f.Close()
-
-	mod := llm.NewGGUFV3(m.Params.ByteOrder)
-	if err := mod.Encode(f, kv, m.Tensors); err != nil {
-		return "", err
-	}
-
-	return f.Name(), nil
+	return llm.NewGGUFV3(m.Params.ByteOrder).Encode(ws, kv, m.Tensors)
 }

convert/mixtral.go

@@ -1,7 +1,7 @@
 package convert
 
 import (
-	"os"
+	"io"
 	"regexp"
 
 	"github.com/ollama/ollama/llm"
@@ -47,7 +47,7 @@ func (m *MixtralModel) LoadVocab() error {
 	return nil
 }
 
-func (m *MixtralModel) WriteGGUF() (string, error) {
+func (m *MixtralModel) WriteGGUF(ws io.WriteSeeker) error {
 	kv := llm.KV{
 		"general.architecture":          "llama",
 		"general.name":                  m.Name,
@@ -81,16 +81,5 @@ func (m *MixtralModel) WriteGGUF() (string, error) {
 		"tokenizer.ggml.add_eos_token":    false,
 	}
 
-	f, err := os.CreateTemp("", "ollama-gguf")
-	if err != nil {
-		return "", err
-	}
-	defer f.Close()
-
-	mod := llm.NewGGUFV3(m.Params.ByteOrder)
-	if err := mod.Encode(f, kv, m.Tensors); err != nil {
-		return "", err
-	}
-
-	return f.Name(), nil
+	return llm.NewGGUFV3(m.Params.ByteOrder).Encode(ws, kv, m.Tensors)
 }

docs/development.md

@@ -51,7 +51,7 @@ Typically the build scripts will auto-detect CUDA, however, if your Linux distro
 or installation approach uses unusual paths, you can specify the location by
 specifying an environment variable `CUDA_LIB_DIR` to the location of the shared
 libraries, and `CUDACXX` to the location of the nvcc compiler. You can customize
-set set of target CUDA architectues by setting `CMAKE_CUDA_ARCHITECTURES` (e.g. "50;60;70")
+a set of target CUDA architectures by setting `CMAKE_CUDA_ARCHITECTURES` (e.g. "50;60;70")
 
 Then generate dependencies:
 
@@ -142,4 +142,4 @@ In addition to the common Windows development tools described above, install AMD
 - [AMD HIP](https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html)
 - [Strawberry Perl](https://strawberryperl.com/)
 
-Lastly, add `ninja.exe` included with MSVC to the system path (e.g. `C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\CommonExtensions\Microsoft\CMake\Ninja`).
+Lastly, add `ninja.exe` included with MSVC to the system path (e.g. `C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\CommonExtensions\Microsoft\CMake\Ninja`).

docs/tutorials/langchainpy.md

@@ -17,10 +17,12 @@ Let's start by asking a simple question that we can get an answer to from the **
 Then we can create a model and ask the question:
 
 ```python
-from langchain.llms import Ollama
-ollama = Ollama(base_url='http://localhost:11434',
-model="llama2")
-print(ollama("why is the sky blue"))
+from langchain_community.llms import Ollama
+ollama = Ollama(
+    base_url='http://localhost:11434',
+    model="llama3"
+)
+print(ollama.invoke("why is the sky blue"))
 ```
 
 Notice that we are defining the model and the base URL for Ollama.

docs/windows.md

@@ -1,47 +1,47 @@
-# Ollama Windows Preview
-
-Welcome to the Ollama Windows preview.
-
-No more WSL required!
-
-Ollama now runs as a native Windows application, including NVIDIA and AMD Radeon GPU support.
-After installing Ollama Windows Preview, Ollama will run in the background and
-the `ollama` command line is available in `cmd`, `powershell` or your favorite
-terminal application. As usual the Ollama [api](./api.md) will be served on
-`http://localhost:11434`.
-
-As this is a preview release, you should expect a few bugs here and there.  If
-you run into a problem you can reach out on
-[Discord](https://discord.gg/ollama), or file an 
-[issue](https://github.com/ollama/ollama/issues).
-Logs will often be helpful in diagnosing the problem (see
-[Troubleshooting](#troubleshooting) below)
-
-## System Requirements
-
-* Windows 10 or newer, Home or Pro
-* NVIDIA 452.39 or newer Drivers if you have an NVIDIA card
-* AMD Radeon Driver https://www.amd.com/en/support if you have a Radeon card
-
-## API Access
-
-Here's a quick example showing API access from `powershell`
-```powershell
-(Invoke-WebRequest -method POST -Body '{"model":"llama2", "prompt":"Why is the sky blue?", "stream": false}' -uri http://localhost:11434/api/generate ).Content | ConvertFrom-json
-```
-
-## Troubleshooting
-
-While we're in preview, `OLLAMA_DEBUG` is always enabled, which adds
-a "view logs" menu item to the app, and increses logging for the GUI app and
-server.
-
-Ollama on Windows stores files in a few different locations.  You can view them in
-the explorer window by hitting `<cmd>+R` and type in:
-- `explorer %LOCALAPPDATA%\Ollama` contains logs, and downloaded updates
-    - *app.log* contains logs from the GUI application
-    - *server.log* contains the server logs
-    - *upgrade.log* contains log output for upgrades
-- `explorer %LOCALAPPDATA%\Programs\Ollama` contains the binaries (The installer adds this to your user PATH)
-- `explorer %HOMEPATH%\.ollama` contains models and configuration
-- `explorer %TEMP%` contains temporary executable files in one or more `ollama*` directories
+# Ollama Windows Preview
+
+Welcome to the Ollama Windows preview.
+
+No more WSL required!
+
+Ollama now runs as a native Windows application, including NVIDIA and AMD Radeon GPU support.
+After installing Ollama Windows Preview, Ollama will run in the background and
+the `ollama` command line is available in `cmd`, `powershell` or your favorite
+terminal application. As usual the Ollama [api](./api.md) will be served on
+`http://localhost:11434`.
+
+As this is a preview release, you should expect a few bugs here and there.  If
+you run into a problem you can reach out on
+[Discord](https://discord.gg/ollama), or file an
+[issue](https://github.com/ollama/ollama/issues).
+Logs will often be helpful in diagnosing the problem (see
+[Troubleshooting](#troubleshooting) below)
+
+## System Requirements
+
+* Windows 10 or newer, Home or Pro
+* NVIDIA 452.39 or newer Drivers if you have an NVIDIA card
+* AMD Radeon Driver https://www.amd.com/en/support if you have a Radeon card
+
+## API Access
+
+Here's a quick example showing API access from `powershell`
+```powershell
+(Invoke-WebRequest -method POST -Body '{"model":"llama2", "prompt":"Why is the sky blue?", "stream": false}' -uri http://localhost:11434/api/generate ).Content | ConvertFrom-json
+```
+
+## Troubleshooting
+
+While we're in preview, `OLLAMA_DEBUG` is always enabled, which adds
+a "view logs" menu item to the app, and increses logging for the GUI app and
+server.
+
+Ollama on Windows stores files in a few different locations.  You can view them in
+the explorer window by hitting `<cmd>+R` and type in:
+- `explorer %LOCALAPPDATA%\Ollama` contains logs, and downloaded updates
+    - *app.log* contains logs from the GUI application
+    - *server.log* contains the server logs
+    - *upgrade.log* contains log output for upgrades
+- `explorer %LOCALAPPDATA%\Programs\Ollama` contains the binaries (The installer adds this to your user PATH)
+- `explorer %HOMEPATH%\.ollama` contains models and configuration
+- `explorer %TEMP%` contains temporary executable files in one or more `ollama*` directories

gpu/assets.go

@@ -40,7 +40,7 @@ func PayloadsDir() (string, error) {
 			}
 
 			var paths []string
-			for _, root := range []string{appExe, cwd} {
+			for _, root := range []string{filepath.Dir(appExe), cwd} {
 				paths = append(paths,
 					filepath.Join(root),
 					filepath.Join(root, "windows-"+runtime.GOARCH),

gpu/gpu_darwin.go

@@ -1,3 +1,5 @@
+//go:build darwin
+
 package gpu
 
 /*
@@ -8,6 +10,12 @@ package gpu
 import "C"
 import (
 	"runtime"
+
+	"github.com/ollama/ollama/format"
+)
+
+const (
+	metalMinimumMemory = 512 * format.MebiByte
 )
 
 func GetGPUInfo() GpuInfoList {
@@ -30,7 +38,7 @@ func GetGPUInfo() GpuInfoList {
 	// TODO is there a way to gather actual allocated video memory? (currentAllocatedSize doesn't work)
 	info.FreeMemory = info.TotalMemory
 
-	info.MinimumMemory = 0
+	info.MinimumMemory = metalMinimumMemory
 	return []GpuInfo{info}
 }
 

integration/utils_test.go

@@ -107,7 +107,7 @@ func startServer(ctx context.Context, ollamaHost string) error {
 
 	if tmp := os.Getenv("OLLAMA_HOST"); tmp != ollamaHost {
 		slog.Info("setting env", "OLLAMA_HOST", ollamaHost)
-		os.Setenv("OLLAMA_HOST", ollamaHost)
+		t.Setenv("OLLAMA_HOST", ollamaHost)
 	}
 
 	slog.Info("starting server", "url", ollamaHost)

llm/ext_server/server.cpp

@@ -1032,7 +1032,7 @@ struct llama_server_context
             slot.has_next_token = false;
         }
 
-        if (!slot.cache_tokens.empty() && result.tok == llama_token_eos(model))
+        if (!slot.cache_tokens.empty() && llama_token_is_eog(model, result.tok))
         {
             slot.stopped_eos = true;
             slot.has_next_token = false;
@@ -1144,12 +1144,15 @@ struct llama_server_context
 
         res.result_json = json
         {
-            {"content",    tkn.text_to_send},
             {"stop",       false},
             {"slot_id",    slot.id},
             {"multimodal", multimodal}
         };
 
+        if (!llama_token_is_eog(model, tkn.tok)) {
+            res.result_json["content"] = tkn.text_to_send;
+        }
+
         if (slot.sparams.n_probs > 0)
         {
             std::vector<completion_token_output> probs_output = {};
@@ -2644,18 +2647,18 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
             if (strncmp(sep, "int:", 4) == 0) {
                 sep += 4;
                 kvo.tag = LLAMA_KV_OVERRIDE_TYPE_INT;
-                kvo.int_value = std::atol(sep);
+                kvo.val_i64 = std::atol(sep);
             } else if (strncmp(sep, "float:", 6) == 0) {
                 sep += 6;
                 kvo.tag = LLAMA_KV_OVERRIDE_TYPE_FLOAT;
-                kvo.float_value = std::atof(sep);
+                kvo.val_f64 = std::atof(sep);
             } else if (strncmp(sep, "bool:", 5) == 0) {
                 sep += 5;
                 kvo.tag = LLAMA_KV_OVERRIDE_TYPE_BOOL;
                 if (std::strcmp(sep, "true") == 0) {
-                    kvo.bool_value = true;
+                    kvo.val_bool = true;
                 } else if (std::strcmp(sep, "false") == 0) {
-                    kvo.bool_value = false;
+                    kvo.val_bool = false;
                 } else {
                     fprintf(stderr, "error: Invalid boolean value for KV override: %s\n", argv[i]);
                     invalid_param = true;

llm/filetype.go

@@ -0,0 +1,140 @@
+package llm
+
+import "fmt"
+
+type fileType uint32
+
+const (
+	fileTypeF32 fileType = iota
+	fileTypeF16
+	fileTypeQ4_0
+	fileTypeQ4_1
+	fileTypeQ4_1_F16
+	fileTypeQ4_2 // unused
+	fileTypeQ4_3 // unused
+	fileTypeQ8_0
+	fileTypeQ5_0
+	fileTypeQ5_1
+	fileTypeQ2_K
+	fileTypeQ3_K_S
+	fileTypeQ3_K_M
+	fileTypeQ3_K_L
+	fileTypeQ4_K_S
+	fileTypeQ4_K_M
+	fileTypeQ5_K_S
+	fileTypeQ5_K_M
+	fileTypeQ6_K
+	fileTypeIQ2_XXS
+	fileTypeIQ2_XS
+	fileTypeQ2_K_S
+	fileTypeQ3_K_XS
+	fileTypeIQ3_XXS
+
+	fileTypeUnknown
+)
+
+func ParseFileType(s string) (fileType, error) {
+	switch s {
+	case "F32":
+		return fileTypeF32, nil
+	case "F16":
+		return fileTypeF16, nil
+	case "Q4_0":
+		return fileTypeQ4_0, nil
+	case "Q4_1":
+		return fileTypeQ4_1, nil
+	case "Q4_1_F16":
+		return fileTypeQ4_1_F16, nil
+	case "Q8_0":
+		return fileTypeQ8_0, nil
+	case "Q5_0":
+		return fileTypeQ5_0, nil
+	case "Q5_1":
+		return fileTypeQ5_1, nil
+	case "Q2_K":
+		return fileTypeQ2_K, nil
+	case "Q3_K_S":
+		return fileTypeQ3_K_S, nil
+	case "Q3_K_M":
+		return fileTypeQ3_K_M, nil
+	case "Q3_K_L":
+		return fileTypeQ3_K_L, nil
+	case "Q4_K_S":
+		return fileTypeQ4_K_S, nil
+	case "Q4_K_M":
+		return fileTypeQ4_K_M, nil
+	case "Q5_K_S":
+		return fileTypeQ5_K_S, nil
+	case "Q5_K_M":
+		return fileTypeQ5_K_M, nil
+	case "Q6_K":
+		return fileTypeQ6_K, nil
+	case "IQ2_XXS":
+		return fileTypeIQ2_XXS, nil
+	case "IQ2_XS":
+		return fileTypeIQ2_XS, nil
+	case "Q2_K_S":
+		return fileTypeQ2_K_S, nil
+	case "Q3_K_XS":
+		return fileTypeQ3_K_XS, nil
+	case "IQ3_XXS":
+		return fileTypeIQ3_XXS, nil
+	default:
+		return fileTypeUnknown, fmt.Errorf("unknown fileType: %s", s)
+	}
+}
+
+func (t fileType) String() string {
+	switch t {
+	case fileTypeF32:
+		return "F32"
+	case fileTypeF16:
+		return "F16"
+	case fileTypeQ4_0:
+		return "Q4_0"
+	case fileTypeQ4_1:
+		return "Q4_1"
+	case fileTypeQ4_1_F16:
+		return "Q4_1_F16"
+	case fileTypeQ8_0:
+		return "Q8_0"
+	case fileTypeQ5_0:
+		return "Q5_0"
+	case fileTypeQ5_1:
+		return "Q5_1"
+	case fileTypeQ2_K:
+		return "Q2_K"
+	case fileTypeQ3_K_S:
+		return "Q3_K_S"
+	case fileTypeQ3_K_M:
+		return "Q3_K_M"
+	case fileTypeQ3_K_L:
+		return "Q3_K_L"
+	case fileTypeQ4_K_S:
+		return "Q4_K_S"
+	case fileTypeQ4_K_M:
+		return "Q4_K_M"
+	case fileTypeQ5_K_S:
+		return "Q5_K_S"
+	case fileTypeQ5_K_M:
+		return "Q5_K_M"
+	case fileTypeQ6_K:
+		return "Q6_K"
+	case fileTypeIQ2_XXS:
+		return "IQ2_XXS"
+	case fileTypeIQ2_XS:
+		return "IQ2_XS"
+	case fileTypeQ2_K_S:
+		return "Q2_K_S"
+	case fileTypeQ3_K_XS:
+		return "Q3_K_XS"
+	case fileTypeIQ3_XXS:
+		return "IQ3_XXS"
+	default:
+		return "unknown"
+	}
+}
+
+func (t fileType) Value() uint32 {
+	return uint32(t)
+}

llm/generate/gen_windows.ps1

@@ -42,7 +42,7 @@ function init_vars {
         "-DLLAMA_NATIVE=off"
         )
     $script:commonCpuDefs = @("-DCMAKE_POSITION_INDEPENDENT_CODE=on")
-    $script:ARCH = "amd64" # arm not yet supported.
+    $script:ARCH = $Env:PROCESSOR_ARCHITECTURE.ToLower()
     $script:DIST_BASE = "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_runners"
     md "$script:DIST_BASE" -ea 0 > $null
     if ($env:CGO_CFLAGS -contains "-g") {
@@ -213,11 +213,11 @@ function build_static() {
     }
 }
 
-function build_cpu() {
+function build_cpu($gen_arch) {
     if ((-not "${env:OLLAMA_SKIP_CPU_GENERATE}" ) -and ((-not "${env:OLLAMA_CPU_TARGET}") -or ("${env:OLLAMA_CPU_TARGET}" -eq "cpu"))) {
         # remaining llama.cpp builds use MSVC 
         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:cmakeDefs = $script:commonCpuDefs + @("-A", $gen_arch, "-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"
@@ -349,11 +349,15 @@ if ($($args.count) -eq 0) {
     git_module_setup
     apply_patches
     build_static
-    build_cpu
-    build_cpu_avx
-    build_cpu_avx2
-    build_cuda
-    build_rocm
+    if ($script:ARCH -eq "arm64") {
+        build_cpu("ARM64")
+    } else { # amd64
+        build_cpu("x64")
+        build_cpu_avx
+        build_cpu_avx2
+        build_cuda
+        build_rocm
+    }
 
     cleanup
     write-host "`ngo generate completed.  LLM runners: $(get-childitem -path $script:DIST_BASE)"

llm/ggla.go

@@ -33,6 +33,7 @@ func (c *containerGGLA) Decode(rs io.ReadSeeker) (model, error) {
 
 type ggla struct {
 	*containerGGLA
+	offset int64
 
 	kv      KV
 	tensors []*Tensor
@@ -53,6 +54,10 @@ func (llm *ggla) Tensors() Tensors {
 	return llm.tensors
 }
 
+func (llm *ggla) Offset() int64 {
+	return llm.offset
+}
+
 func (llm *ggla) decode(rs io.ReadSeeker) error {
 	var r uint32
 	if err := binary.Read(rs, binary.LittleEndian, &r); err != nil {
@@ -66,6 +71,13 @@ func (llm *ggla) decode(rs io.ReadSeeker) error {
 	}
 	llm.kv["alpha"] = alpha
 
+	offset, err := rs.Seek(0, io.SeekCurrent)
+	if err != nil {
+		return err
+	}
+
+	llm.offset = offset
+
 	for {
 		var dims uint32
 		if err := binary.Read(rs, binary.LittleEndian, &dims); err != nil {

llm/ggml.go

@@ -13,85 +13,10 @@ type GGML struct {
 	model
 }
 
-const (
-	fileTypeF32 uint32 = iota
-	fileTypeF16
-	fileTypeQ4_0
-	fileTypeQ4_1
-	fileTypeQ4_1_F16
-	fileTypeQ8_0 uint32 = iota + 2
-	fileTypeQ5_0
-	fileTypeQ5_1
-	fileTypeQ2_K
-	fileTypeQ3_K_S
-	fileTypeQ3_K_M
-	fileTypeQ3_K_L
-	fileTypeQ4_K_S
-	fileTypeQ4_K_M
-	fileTypeQ5_K_S
-	fileTypeQ5_K_M
-	fileTypeQ6_K
-	fileTypeIQ2_XXS
-	fileTypeIQ2_XS
-	fileTypeQ2_K_S
-	fileTypeQ3_K_XS
-	fileTypeIQ3_XXS
-)
-
-func fileType(fileType uint32) string {
-	switch fileType {
-	case fileTypeF32:
-		return "F32"
-	case fileTypeF16:
-		return "F16"
-	case fileTypeQ4_0:
-		return "Q4_0"
-	case fileTypeQ4_1:
-		return "Q4_1"
-	case fileTypeQ4_1_F16:
-		return "Q4_1_F16"
-	case fileTypeQ8_0:
-		return "Q8_0"
-	case fileTypeQ5_0:
-		return "Q5_0"
-	case fileTypeQ5_1:
-		return "Q5_1"
-	case fileTypeQ2_K:
-		return "Q2_K"
-	case fileTypeQ3_K_S:
-		return "Q3_K_S"
-	case fileTypeQ3_K_M:
-		return "Q3_K_M"
-	case fileTypeQ3_K_L:
-		return "Q3_K_L"
-	case fileTypeQ4_K_S:
-		return "Q4_K_S"
-	case fileTypeQ4_K_M:
-		return "Q4_K_M"
-	case fileTypeQ5_K_S:
-		return "Q5_K_S"
-	case fileTypeQ5_K_M:
-		return "Q5_K_M"
-	case fileTypeQ6_K:
-		return "Q6_K"
-	case fileTypeIQ2_XXS:
-		return "IQ2_XXS"
-	case fileTypeIQ2_XS:
-		return "IQ2_XS"
-	case fileTypeQ2_K_S:
-		return "Q2_K_S"
-	case fileTypeQ3_K_XS:
-		return "Q3_K_XS"
-	case fileTypeIQ3_XXS:
-		return "IQ3_XXS"
-	default:
-		return "unknown"
-	}
-}
-
 type model interface {
 	KV() KV
 	Tensors() Tensors
+	Offset() int64
 }
 
 type KV map[string]any
@@ -123,7 +48,7 @@ func (kv KV) ParameterCount() uint64 {
 
 func (kv KV) FileType() string {
 	if u64 := kv.u64("general.file_type"); u64 > 0 {
-		return fileType(uint32(u64))
+		return fileType(uint32(u64)).String()
 	}
 
 	return "unknown"
@@ -286,6 +211,23 @@ const (
 
 var ErrUnsupportedFormat = errors.New("unsupported model format")
 
+func DetectGGMLType(b []byte) string {
+	switch binary.LittleEndian.Uint32(b[:4]) {
+	case FILE_MAGIC_GGML:
+		return "ggml"
+	case FILE_MAGIC_GGMF:
+		return "ggmf"
+	case FILE_MAGIC_GGJT:
+		return "ggjt"
+	case FILE_MAGIC_GGLA:
+		return "ggla"
+	case FILE_MAGIC_GGUF_LE, FILE_MAGIC_GGUF_BE:
+		return "gguf"
+	default:
+		return ""
+	}
+}
+
 func DecodeGGML(rs io.ReadSeeker) (*GGML, int64, error) {
 	var magic uint32
 	if err := binary.Read(rs, binary.LittleEndian, &magic); err != nil {

llm/gguf.go

@@ -55,7 +55,7 @@ func (c *containerGGUF) Decode(rs io.ReadSeeker) (model, error) {
 
 	model := newGGUF(c)
 	slog.Debug(fmt.Sprintf("model = %#v", model))
-	if err := model.Decode(rs); err != nil {
+	if err := model.decode(rs); err != nil {
 		return nil, err
 	}
 
@@ -90,6 +90,7 @@ const (
 
 type gguf struct {
 	*containerGGUF
+	offset int64
 
 	kv      KV
 	tensors []*Tensor
@@ -116,6 +117,10 @@ func (llm *gguf) Tensors() Tensors {
 	return llm.tensors
 }
 
+func (llm *gguf) Offset() int64 {
+	return llm.offset
+}
+
 func (llm *gguf) numTensor() uint64 {
 	switch llm.Version {
 	case 1:
@@ -138,7 +143,7 @@ func (llm *gguf) numKV() uint64 {
 	}
 }
 
-func (llm *gguf) Decode(rs io.ReadSeeker) error {
+func (llm *gguf) decode(rs io.ReadSeeker) error {
 	// decode key-values
 	for i := 0; uint64(i) < llm.numKV(); i++ {
 		k, err := readGGUFString(llm, rs)
@@ -250,6 +255,8 @@ func (llm *gguf) Decode(rs io.ReadSeeker) error {
 		return err
 	}
 
+	llm.offset = offset + padding
+
 	for _, tensor := range llm.tensors {
 		if _, err := rs.Seek(int64(tensor.size()), io.SeekCurrent); err != nil {
 			return err

llm/llm.go

@@ -4,6 +4,7 @@ package llm
 // #cgo darwin,arm64 LDFLAGS: ${SRCDIR}/build/darwin/arm64_static/libllama.a -lstdc++
 // #cgo darwin,amd64 LDFLAGS: ${SRCDIR}/build/darwin/x86_64_static/libllama.a -lstdc++
 // #cgo windows,amd64 LDFLAGS: ${SRCDIR}/build/windows/amd64_static/libllama.a -static -lstdc++
+// #cgo windows,arm64 LDFLAGS: ${SRCDIR}/build/windows/arm64_static/libllama.a -static -lstdc++
 // #cgo linux,amd64 LDFLAGS: ${SRCDIR}/build/linux/x86_64_static/libllama.a -lstdc++
 // #cgo linux,arm64 LDFLAGS: ${SRCDIR}/build/linux/arm64_static/libllama.a -lstdc++
 // #include <stdlib.h>
@@ -19,7 +20,7 @@ func SystemInfo() string {
 	return C.GoString(C.llama_print_system_info())
 }
 
-func Quantize(infile, outfile, filetype string) error {
+func Quantize(infile, outfile string, ftype fileType) error {
 	cinfile := C.CString(infile)
 	defer C.free(unsafe.Pointer(cinfile))
 
@@ -28,58 +29,10 @@ func Quantize(infile, outfile, filetype string) error {
 
 	params := C.llama_model_quantize_default_params()
 	params.nthread = -1
+	params.ftype = ftype.Value()
 
-	switch filetype {
-	case "F32":
-		params.ftype = fileTypeF32
-	case "F16":
-		params.ftype = fileTypeF16
-	case "Q4_0":
-		params.ftype = fileTypeQ4_0
-	case "Q4_1":
-		params.ftype = fileTypeQ4_1
-	case "Q4_1_F16":
-		params.ftype = fileTypeQ4_1_F16
-	case "Q8_0":
-		params.ftype = fileTypeQ8_0
-	case "Q5_0":
-		params.ftype = fileTypeQ5_0
-	case "Q5_1":
-		params.ftype = fileTypeQ5_1
-	case "Q2_K":
-		params.ftype = fileTypeQ2_K
-	case "Q3_K_S":
-		params.ftype = fileTypeQ3_K_S
-	case "Q3_K_M":
-		params.ftype = fileTypeQ3_K_M
-	case "Q3_K_L":
-		params.ftype = fileTypeQ3_K_L
-	case "Q4_K_S":
-		params.ftype = fileTypeQ4_K_S
-	case "Q4_K_M":
-		params.ftype = fileTypeQ4_K_M
-	case "Q5_K_S":
-		params.ftype = fileTypeQ5_K_S
-	case "Q5_K_M":
-		params.ftype = fileTypeQ5_K_M
-	case "Q6_K":
-		params.ftype = fileTypeQ6_K
-	case "IQ2_XXS":
-		params.ftype = fileTypeIQ2_XXS
-	case "IQ2_XS":
-		params.ftype = fileTypeIQ2_XS
-	case "Q2_K_S":
-		params.ftype = fileTypeQ2_K_S
-	case "Q3_K_XS":
-		params.ftype = fileTypeQ3_K_XS
-	case "IQ3_XXS":
-		params.ftype = fileTypeIQ3_XXS
-	default:
-		return fmt.Errorf("unknown filetype: %s", filetype)
-	}
-
-	if retval := C.llama_model_quantize(cinfile, coutfile, &params); retval != 0 {
-		return fmt.Errorf("llama_model_quantize: %d", retval)
+	if rc := C.llama_model_quantize(cinfile, coutfile, &params); rc != 0 {
+		return fmt.Errorf("llama_model_quantize: %d", rc)
 	}
 
 	return nil

llm/memory.go

@@ -88,6 +88,11 @@ func EstimateGPULayers(gpus []gpu.GpuInfo, ggml *GGML, projectors []string, opts
 	graphFullOffload *= uint64(len(gpus))
 	graphPartialOffload *= uint64(len(gpus))
 
+	// on metal there's no partial offload overhead
+	if gpus[0].Library == "metal" {
+		graphPartialOffload = graphFullOffload
+	}
+
 	// memoryRequiredTotal represents the memory required for full GPU offloading (all layers)
 	memoryRequiredTotal := memoryMinimum + graphFullOffload
 

llm/server.go

@@ -73,8 +73,7 @@ func LoadModel(model string) (*GGML, error) {
 func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, projectors []string, opts api.Options) (LlamaServer, error) {
 	var err error
 	if opts.NumCtx > int(ggml.KV().ContextLength()) {
-		slog.Warn("requested context length is greater than model max context length", "requested", opts.NumCtx, "model", ggml.KV().ContextLength())
-		opts.NumCtx = int(ggml.KV().ContextLength())
+		slog.Warn("requested context length is greater than the model's training context window size", "requested", opts.NumCtx, "training size", ggml.KV().ContextLength())
 	}
 
 	if opts.NumCtx < 4 {
@@ -301,12 +300,6 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 			continue
 		}
 
-		// reap subprocess when it exits
-		go func() {
-			// Exit status managed via getServerStatus
-			_ = s.cmd.Wait()
-		}()
-
 		// TODO - make sure this is all wired up correctly
 		// if err = s.WaitUntilRunning(); err != nil {
 		// 	slog.Error("error starting llama server", "server", servers[i], "error", err)
@@ -900,7 +893,13 @@ func (s *llmServer) Detokenize(ctx context.Context, tokens []int) (string, error
 func (s *llmServer) Close() error {
 	if s.cmd != nil {
 		slog.Debug("stopping llama server")
-		return s.cmd.Process.Kill()
+		if err := s.cmd.Process.Kill(); err != nil {
+			return err
+		}
+
+		_ = s.cmd.Wait()
+
+		slog.Debug("llama server stopped")
 	}
 
 	return nil

macapp/.eslintrc.json

@@ -0,0 +1,16 @@
+{
+  "env": {
+    "browser": true,
+    "es6": true,
+    "node": true
+  },
+  "extends": [
+    "eslint:recommended",
+    "plugin:@typescript-eslint/eslint-recommended",
+    "plugin:@typescript-eslint/recommended",
+    "plugin:import/recommended",
+    "plugin:import/electron",
+    "plugin:import/typescript"
+  ],
+  "parser": "@typescript-eslint/parser"
+}

macapp/.gitignore

@@ -0,0 +1,92 @@
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+lerna-debug.log*
+
+# Diagnostic reports (https://nodejs.org/api/report.html)
+report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json
+
+# Runtime data
+pids
+*.pid
+*.seed
+*.pid.lock
+.DS_Store
+
+# Directory for instrumented libs generated by jscoverage/JSCover
+lib-cov
+
+# Coverage directory used by tools like istanbul
+coverage
+*.lcov
+
+# nyc test coverage
+.nyc_output
+
+# node-waf configuration
+.lock-wscript
+
+# Compiled binary addons (https://nodejs.org/api/addons.html)
+build/Release
+
+# Dependency directories
+node_modules/
+jspm_packages/
+
+# TypeScript v1 declaration files
+typings/
+
+# TypeScript cache
+*.tsbuildinfo
+
+# Optional npm cache directory
+.npm
+
+# Optional eslint cache
+.eslintcache
+
+# Optional REPL history
+.node_repl_history
+
+# Output of 'npm pack'
+*.tgz
+
+# Yarn Integrity file
+.yarn-integrity
+
+# dotenv environment variables file
+.env
+.env.test
+
+# parcel-bundler cache (https://parceljs.org/)
+.cache
+
+# next.js build output
+.next
+
+# nuxt.js build output
+.nuxt
+
+# vuepress build output
+.vuepress/dist
+
+# Serverless directories
+.serverless/
+
+# FuseBox cache
+.fusebox/
+
+# DynamoDB Local files
+.dynamodb/
+
+# Webpack
+.webpack/
+
+# Vite
+.vite/
+
+# Electron-Forge
+out/

macapp/README.md

@@ -0,0 +1,21 @@
+# Desktop
+
+This app builds upon Ollama to provide a desktop experience for running models.
+
+## Developing
+
+First, build the `ollama` binary:
+
+```
+cd ..
+go build .
+```
+
+Then run the desktop app with `npm start`:
+
+```
+cd macapp
+npm install
+npm start
+```
+

macapp/forge.config.ts

@@ -0,0 +1,78 @@
+import type { ForgeConfig } from '@electron-forge/shared-types'
+import { MakerSquirrel } from '@electron-forge/maker-squirrel'
+import { MakerZIP } from '@electron-forge/maker-zip'
+import { PublisherGithub } from '@electron-forge/publisher-github'
+import { AutoUnpackNativesPlugin } from '@electron-forge/plugin-auto-unpack-natives'
+import { WebpackPlugin } from '@electron-forge/plugin-webpack'
+import * as path from 'path'
+import * as fs from 'fs'
+
+import { mainConfig } from './webpack.main.config'
+import { rendererConfig } from './webpack.renderer.config'
+
+const packageJson = JSON.parse(fs.readFileSync(path.resolve(__dirname, './package.json'), 'utf8'))
+
+const config: ForgeConfig = {
+  packagerConfig: {
+    appVersion: process.env.VERSION || packageJson.version,
+    asar: true,
+    icon: './assets/icon.icns',
+    extraResource: [
+      '../dist/ollama',
+      path.join(__dirname, './assets/iconTemplate.png'),
+      path.join(__dirname, './assets/iconTemplate@2x.png'),
+      path.join(__dirname, './assets/iconUpdateTemplate.png'),
+      path.join(__dirname, './assets/iconUpdateTemplate@2x.png'),
+      path.join(__dirname, './assets/iconDarkTemplate.png'),
+      path.join(__dirname, './assets/iconDarkTemplate@2x.png'),
+      path.join(__dirname, './assets/iconDarkUpdateTemplate.png'),
+      path.join(__dirname, './assets/iconDarkUpdateTemplate@2x.png'),
+    ],
+    ...(process.env.SIGN
+      ? {
+          osxSign: {
+            identity: process.env.APPLE_IDENTITY,
+          },
+          osxNotarize: {
+            tool: 'notarytool',
+            appleId: process.env.APPLE_ID || '',
+            appleIdPassword: process.env.APPLE_PASSWORD || '',
+            teamId: process.env.APPLE_TEAM_ID || '',
+          },
+        }
+      : {}),
+    osxUniversal: {
+      x64ArchFiles: '**/ollama',
+    },
+  },
+  rebuildConfig: {},
+  makers: [new MakerSquirrel({}), new MakerZIP({}, ['darwin'])],
+  hooks: {
+    readPackageJson: async (_, packageJson) => {
+      return { ...packageJson, version: process.env.VERSION || packageJson.version }
+    },
+  },
+  plugins: [
+    new AutoUnpackNativesPlugin({}),
+    new WebpackPlugin({
+      mainConfig,
+      devContentSecurityPolicy: `default-src * 'unsafe-eval' 'unsafe-inline'; img-src data: 'self'`,
+      renderer: {
+        config: rendererConfig,
+        nodeIntegration: true,
+        entryPoints: [
+          {
+            html: './src/index.html',
+            js: './src/renderer.tsx',
+            name: 'main_window',
+            preload: {
+              js: './src/preload.ts',
+            },
+          },
+        ],
+      },
+    }),
+  ],
+}
+
+export default config

macapp/package.json

@@ -0,0 +1,84 @@
+{
+  "name": "ollama",
+  "productName": "Ollama",
+  "version": "0.0.0",
+  "description": "ollama",
+  "main": ".webpack/main",
+  "scripts": {
+    "start": "electron-forge start",
+    "package": "electron-forge package --arch universal",
+    "package:sign": "SIGN=1 electron-forge package --arch universal",
+    "make": "electron-forge make --arch universal",
+    "make:sign": "SIGN=1 electron-forge make --arch universal",
+    "publish": "SIGN=1 electron-forge publish",
+    "lint": "eslint --ext .ts,.tsx .",
+    "format": "prettier --check . --ignore-path .gitignore",
+    "format:fix": "prettier --write . --ignore-path .gitignore"
+  },
+  "keywords": [],
+  "author": {
+    "name": "Jeffrey Morgan",
+    "email": "jmorganca@gmail.com"
+  },
+  "license": "MIT",
+  "devDependencies": {
+    "@babel/core": "^7.22.5",
+    "@babel/preset-react": "^7.22.5",
+    "@electron-forge/cli": "^6.2.1",
+    "@electron-forge/maker-deb": "^6.2.1",
+    "@electron-forge/maker-rpm": "^6.2.1",
+    "@electron-forge/maker-squirrel": "^6.2.1",
+    "@electron-forge/maker-zip": "^6.2.1",
+    "@electron-forge/plugin-auto-unpack-natives": "^6.2.1",
+    "@electron-forge/plugin-webpack": "^6.2.1",
+    "@electron-forge/publisher-github": "^6.2.1",
+    "@electron/universal": "^1.4.1",
+    "@svgr/webpack": "^8.0.1",
+    "@types/chmodr": "^1.0.0",
+    "@types/node": "^20.4.0",
+    "@types/react": "^18.2.14",
+    "@types/react-dom": "^18.2.6",
+    "@types/uuid": "^9.0.2",
+    "@typescript-eslint/eslint-plugin": "^5.60.0",
+    "@typescript-eslint/parser": "^5.60.0",
+    "@vercel/webpack-asset-relocator-loader": "^1.7.3",
+    "babel-loader": "^9.1.2",
+    "chmodr": "^1.2.0",
+    "copy-webpack-plugin": "^11.0.0",
+    "css-loader": "^6.8.1",
+    "electron": "25.9.2",
+    "eslint": "^8.43.0",
+    "eslint-plugin-import": "^2.27.5",
+    "fork-ts-checker-webpack-plugin": "^7.3.0",
+    "node-loader": "^2.0.0",
+    "postcss": "^8.4.24",
+    "postcss-import": "^15.1.0",
+    "postcss-loader": "^7.3.3",
+    "postcss-preset-env": "^8.5.1",
+    "prettier": "^2.8.8",
+    "prettier-plugin-tailwindcss": "^0.3.0",
+    "style-loader": "^3.3.3",
+    "svg-inline-loader": "^0.8.2",
+    "tailwindcss": "^3.3.2",
+    "ts-loader": "^9.4.3",
+    "ts-node": "^10.9.1",
+    "typescript": "~4.5.4",
+    "url-loader": "^4.1.1",
+    "webpack": "^5.88.0",
+    "webpack-cli": "^5.1.4",
+    "webpack-dev-server": "^4.15.1"
+  },
+  "dependencies": {
+    "@electron/remote": "^2.0.10",
+    "@heroicons/react": "^2.0.18",
+    "@segment/analytics-node": "^1.0.0",
+    "copy-to-clipboard": "^3.3.3",
+    "electron-squirrel-startup": "^1.0.0",
+    "electron-store": "^8.1.0",
+    "react": "^18.2.0",
+    "react-dom": "^18.2.0",
+    "uuid": "^9.0.0",
+    "winston": "^3.10.0",
+    "winston-daily-rotate-file": "^4.7.1"
+  }
+}

macapp/postcss.config.js

@@ -0,0 +1,7 @@
+module.exports = {
+  plugins: {
+    'postcss-import': {},
+    tailwindcss: {},
+    autoprefixer: {},
+  },
+}

macapp/src/app.css

@@ -0,0 +1,34 @@
+@tailwind base;
+@tailwind components;
+@tailwind utilities;
+
+html,
+body {
+  background: transparent;
+}
+
+.drag {
+  -webkit-app-region: drag;
+}
+
+.no-drag {
+  -webkit-app-region: no-drag;
+}
+
+.blink {
+  -webkit-animation: 1s blink step-end infinite;
+  -moz-animation: 1s blink step-end infinite;
+  -ms-animation: 1s blink step-end infinite;
+  -o-animation: 1s blink step-end infinite;
+  animation: 1s blink step-end infinite;
+}
+
+@keyframes blink {
+  from,
+  to {
+    color: transparent;
+  }
+  50% {
+    color: black;
+  }
+}

macapp/src/app.tsx

@@ -0,0 +1,122 @@
+import { useState } from 'react'
+import copy from 'copy-to-clipboard'
+import { CheckIcon, DocumentDuplicateIcon } from '@heroicons/react/24/outline'
+import Store from 'electron-store'
+import { getCurrentWindow, app } from '@electron/remote'
+
+import { install } from './install'
+import OllamaIcon from './ollama.svg'
+
+const store = new Store()
+
+enum Step {
+  WELCOME = 0,
+  CLI,
+  FINISH,
+}
+
+export default function () {
+  const [step, setStep] = useState<Step>(Step.WELCOME)
+  const [commandCopied, setCommandCopied] = useState<boolean>(false)
+
+  const command = 'ollama run llama3'
+
+  return (
+    <div className='drag'>
+      <div className='mx-auto flex min-h-screen w-full flex-col justify-between bg-white px-4 pt-16'>
+        {step === Step.WELCOME && (
+          <>
+            <div className='mx-auto text-center'>
+              <h1 className='mb-6 mt-4 text-2xl tracking-tight text-gray-900'>Welcome to Ollama</h1>
+              <p className='mx-auto w-[65%] text-sm text-gray-400'>
+                Let's get you up and running with your own large language models.
+              </p>
+              <button
+                onClick={() => setStep(Step.CLI)}
+                className='no-drag rounded-dm mx-auto my-8 w-[40%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
+              >
+                Next
+              </button>
+            </div>
+            <div className='mx-auto'>
+              <OllamaIcon />
+            </div>
+          </>
+        )}
+        {step === Step.CLI && (
+          <>
+            <div className='mx-auto flex flex-col space-y-28 text-center'>
+              <h1 className='mt-4 text-2xl tracking-tight text-gray-900'>Install the command line</h1>
+              <pre className='mx-auto text-4xl text-gray-400'>&gt; ollama</pre>
+              <div className='mx-auto'>
+                <button
+                  onClick={async () => {
+                    try {
+                      await install()
+                      setStep(Step.FINISH)
+                    } catch (e) {
+                      console.error('could not install: ', e)
+                    } finally {
+                      getCurrentWindow().show()
+                      getCurrentWindow().focus()
+                    }
+                  }}
+                  className='no-drag rounded-dm mx-auto w-[60%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
+                >
+                  Install
+                </button>
+                <p className='mx-auto my-4 w-[70%] text-xs text-gray-400'>
+                  You will be prompted for administrator access
+                </p>
+              </div>
+            </div>
+          </>
+        )}
+        {step === Step.FINISH && (
+          <>
+            <div className='mx-auto flex flex-col space-y-20 text-center'>
+              <h1 className='mt-4 text-2xl tracking-tight text-gray-900'>Run your first model</h1>
+              <div className='flex flex-col'>
+                <div className='group relative flex items-center'>
+                  <pre className='language-none text-2xs w-full rounded-md bg-gray-100 px-4 py-3 text-start leading-normal'>
+                    {command}
+                  </pre>
+                  <button
+                    className={`no-drag absolute right-[5px] px-2 py-2 ${
+                      commandCopied
+                        ? 'text-gray-900 opacity-100 hover:cursor-auto'
+                        : 'text-gray-200 opacity-50 hover:cursor-pointer'
+                    } hover:font-bold hover:text-gray-900 group-hover:opacity-100`}
+                    onClick={() => {
+                      copy(command)
+                      setCommandCopied(true)
+                      setTimeout(() => setCommandCopied(false), 3000)
+                    }}
+                  >
+                    {commandCopied ? (
+                      <CheckIcon className='h-4 w-4 font-bold text-gray-500' />
+                    ) : (
+                      <DocumentDuplicateIcon className='h-4 w-4 text-gray-500' />
+                    )}
+                  </button>
+                </div>
+                <p className='mx-auto my-4 w-[70%] text-xs text-gray-400'>
+                  Run this command in your favorite terminal.
+                </p>
+              </div>
+              <button
+                onClick={() => {
+                  store.set('first-time-run', true)
+                  window.close()
+                }}
+                className='no-drag rounded-dm mx-auto w-[60%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
+              >
+                Finish
+              </button>
+            </div>
+          </>
+        )}
+      </div>
+    </div>
+  )
+}

macapp/src/declarations.d.ts

@@ -0,0 +1,4 @@
+declare module '*.svg' {
+  const content: string
+  export default content
+}

macapp/src/index.html

@@ -0,0 +1,9 @@
+<!DOCTYPE html>
+<html>
+  <head>
+    <meta charset="UTF-8" />
+  </head>
+  <body>
+    <div id="app"></div>
+  </body>
+</html>

macapp/src/index.ts

@@ -0,0 +1,302 @@
+import { spawn, ChildProcess } from 'child_process'
+import { app, autoUpdater, dialog, Tray, Menu, BrowserWindow, MenuItemConstructorOptions, nativeTheme } from 'electron'
+import Store from 'electron-store'
+import winston from 'winston'
+import 'winston-daily-rotate-file'
+import * as path from 'path'
+
+import { v4 as uuidv4 } from 'uuid'
+import { installed } from './install'
+
+require('@electron/remote/main').initialize()
+
+if (require('electron-squirrel-startup')) {
+  app.quit()
+}
+
+const store = new Store()
+
+let welcomeWindow: BrowserWindow | null = null
+
+declare const MAIN_WINDOW_WEBPACK_ENTRY: string
+
+const logger = winston.createLogger({
+  transports: [
+    new winston.transports.Console(),
+    new winston.transports.File({
+      filename: path.join(app.getPath('home'), '.ollama', 'logs', 'server.log'),
+      maxsize: 1024 * 1024 * 20,
+      maxFiles: 5,
+    }),
+  ],
+  format: winston.format.printf(info => info.message),
+})
+
+app.on('ready', () => {
+  const gotTheLock = app.requestSingleInstanceLock()
+  if (!gotTheLock) {
+    app.exit(0)
+    return
+  }
+
+  app.on('second-instance', () => {
+    if (app.hasSingleInstanceLock()) {
+      app.releaseSingleInstanceLock()
+    }
+
+    if (proc) {
+      proc.off('exit', restart)
+      proc.kill()
+    }
+
+    app.exit(0)
+  })
+
+  app.focus({ steal: true })
+
+  init()
+})
+
+function firstRunWindow() {
+  // Create the browser window.
+  welcomeWindow = new BrowserWindow({
+    width: 400,
+    height: 500,
+    frame: false,
+    fullscreenable: false,
+    resizable: false,
+    movable: true,
+    show: false,
+    webPreferences: {
+      nodeIntegration: true,
+      contextIsolation: false,
+    },
+  })
+
+  require('@electron/remote/main').enable(welcomeWindow.webContents)
+
+  welcomeWindow.loadURL(MAIN_WINDOW_WEBPACK_ENTRY)
+  welcomeWindow.on('ready-to-show', () => welcomeWindow.show())
+  welcomeWindow.on('closed', () => {
+    if (process.platform === 'darwin') {
+      app.dock.hide()
+    }
+  })
+}
+
+let tray: Tray | null = null
+let updateAvailable = false
+const assetPath = app.isPackaged ? process.resourcesPath : path.join(__dirname, '..', '..', 'assets')
+
+function trayIconPath() {
+  return nativeTheme.shouldUseDarkColors
+    ? updateAvailable
+      ? path.join(assetPath, 'iconDarkUpdateTemplate.png')
+      : path.join(assetPath, 'iconDarkTemplate.png')
+    : updateAvailable
+    ? path.join(assetPath, 'iconUpdateTemplate.png')
+    : path.join(assetPath, 'iconTemplate.png')
+}
+
+function updateTrayIcon() {
+  if (tray) {
+    tray.setImage(trayIconPath())
+  }
+}
+
+function updateTray() {
+  const updateItems: MenuItemConstructorOptions[] = [
+    { label: 'An update is available', enabled: false },
+    {
+      label: 'Restart to update',
+      click: () => autoUpdater.quitAndInstall(),
+    },
+    { type: 'separator' },
+  ]
+
+  const menu = Menu.buildFromTemplate([
+    ...(updateAvailable ? updateItems : []),
+    { role: 'quit', label: 'Quit Ollama', accelerator: 'Command+Q' },
+  ])
+
+  if (!tray) {
+    tray = new Tray(trayIconPath())
+  }
+
+  tray.setToolTip(updateAvailable ? 'An update is available' : 'Ollama')
+  tray.setContextMenu(menu)
+  tray.setImage(trayIconPath())
+
+  nativeTheme.off('updated', updateTrayIcon)
+  nativeTheme.on('updated', updateTrayIcon)
+}
+
+let proc: ChildProcess = null
+
+function server() {
+  const binary = app.isPackaged
+    ? path.join(process.resourcesPath, 'ollama')
+    : path.resolve(process.cwd(), '..', 'ollama')
+
+  proc = spawn(binary, ['serve'])
+
+  proc.stdout.on('data', data => {
+    logger.info(data.toString().trim())
+  })
+
+  proc.stderr.on('data', data => {
+    logger.error(data.toString().trim())
+  })
+
+  proc.on('exit', restart)
+}
+
+function restart() {
+  setTimeout(server, 1000)
+}
+
+app.on('before-quit', () => {
+  if (proc) {
+    proc.off('exit', restart)
+    proc.kill('SIGINT') // send SIGINT signal to the server, which also stops any loaded llms
+  }
+})
+
+const updateURL = `https://ollama.ai/api/update?os=${process.platform}&arch=${
+  process.arch
+}&version=${app.getVersion()}&id=${id()}`
+
+let latest = ''
+async function isNewReleaseAvailable() {
+  try {
+    const response = await fetch(updateURL)
+
+    if (!response.ok) {
+      return false
+    }
+
+    if (response.status === 204) {
+      return false
+    }
+
+    const data = await response.json()
+
+    const url = data?.url
+    if (!url) {
+      return false
+    }
+
+    if (latest === url) {
+      return false
+    }
+
+    latest = url
+
+    return true
+  } catch (error) {
+    logger.error(`update check failed - ${error}`)
+    return false
+  }
+}
+
+async function checkUpdate() {
+  const available = await isNewReleaseAvailable()
+  if (available) {
+    logger.info('checking for update')
+    autoUpdater.checkForUpdates()
+  }
+}
+
+function init() {
+  if (app.isPackaged) {
+    checkUpdate()
+    setInterval(() => {
+      checkUpdate()
+    }, 60 * 60 * 1000)
+  }
+
+  updateTray()
+
+  if (process.platform === 'darwin') {
+    if (app.isPackaged) {
+      if (!app.isInApplicationsFolder()) {
+        const chosen = dialog.showMessageBoxSync({
+          type: 'question',
+          buttons: ['Move to Applications', 'Do Not Move'],
+          message: 'Ollama works best when run from the Applications directory.',
+          defaultId: 0,
+          cancelId: 1,
+        })
+
+        if (chosen === 0) {
+          try {
+            app.moveToApplicationsFolder({
+              conflictHandler: conflictType => {
+                if (conflictType === 'existsAndRunning') {
+                  dialog.showMessageBoxSync({
+                    type: 'info',
+                    message: 'Cannot move to Applications directory',
+                    detail:
+                      'Another version of Ollama is currently running from your Applications directory. Close it first and try again.',
+                  })
+                }
+                return true
+              },
+            })
+            return
+          } catch (e) {
+            logger.error(`[Move to Applications] Failed to move to applications folder - ${e.message}}`)
+          }
+        }
+      }
+    }
+  }
+
+  server()
+
+  if (store.get('first-time-run') && installed()) {
+    if (process.platform === 'darwin') {
+      app.dock.hide()
+    }
+
+    app.setLoginItemSettings({ openAtLogin: app.getLoginItemSettings().openAtLogin })
+    return
+  }
+
+  // This is the first run or the CLI is no longer installed
+  app.setLoginItemSettings({ openAtLogin: true })
+  firstRunWindow()
+}
+
+// Quit when all windows are closed, except on macOS. There, it's common
+// for applications and their menu bar to stay active until the user quits
+// explicitly with Cmd + Q.
+app.on('window-all-closed', () => {
+  if (process.platform !== 'darwin') {
+    app.quit()
+  }
+})
+
+function id(): string {
+  const id = store.get('id') as string
+
+  if (id) {
+    return id
+  }
+
+  const uuid = uuidv4()
+  store.set('id', uuid)
+  return uuid
+}
+
+autoUpdater.setFeedURL({ url: updateURL })
+
+autoUpdater.on('error', e => {
+  logger.error(`update check failed - ${e.message}`)
+  console.error(`update check failed - ${e.message}`)
+})
+
+autoUpdater.on('update-downloaded', () => {
+  updateAvailable = true
+  updateTray()
+})

macapp/src/install.ts

@@ -0,0 +1,21 @@
+import * as fs from 'fs'
+import { exec as cbExec } from 'child_process'
+import * as path from 'path'
+import { promisify } from 'util'
+
+const app = process && process.type === 'renderer' ? require('@electron/remote').app : require('electron').app
+const ollama = app.isPackaged ? path.join(process.resourcesPath, 'ollama') : path.resolve(process.cwd(), '..', 'ollama')
+const exec = promisify(cbExec)
+const symlinkPath = '/usr/local/bin/ollama'
+
+export function installed() {
+  return fs.existsSync(symlinkPath) && fs.readlinkSync(symlinkPath) === ollama
+}
+
+export async function install() {
+  const command = `do shell script "mkdir -p ${path.dirname(
+    symlinkPath
+  )} && ln -F -s \\"${ollama}\\" \\"${symlinkPath}\\"" with administrator privileges`
+
+  await exec(`osascript -e '${command}'`)
+}

macapp/src/renderer.tsx

@@ -0,0 +1,7 @@
+import App from './app'
+import './app.css'
+import { createRoot } from 'react-dom/client'
+
+const container = document.getElementById('app')
+const root = createRoot(container)
+root.render(<App />)

macapp/tailwind.config.js

@@ -0,0 +1,6 @@
+/** @type {import('tailwindcss').Config} */
+module.exports = {
+  content: ['./src/**/*.{js,ts,jsx,tsx,mdx}'],
+  theme: {},
+  plugins: [],
+}

macapp/tsconfig.json

@@ -0,0 +1,20 @@
+{
+  "compilerOptions": {
+    "target": "ES6",
+    "allowJs": true,
+    "module": "commonjs",
+    "skipLibCheck": true,
+    "esModuleInterop": true,
+    "noImplicitAny": true,
+    "sourceMap": true,
+    "baseUrl": ".",
+    "outDir": "dist",
+    "moduleResolution": "node",
+    "resolveJsonModule": true,
+    "paths": {
+      "*": ["node_modules/*"]
+    },
+    "jsx": "react-jsx"
+  },
+  "include": ["src/**/*"]
+}

macapp/webpack.main.config.ts

@@ -0,0 +1,20 @@
+import type { Configuration } from 'webpack'
+
+import { rules } from './webpack.rules'
+import { plugins } from './webpack.plugins'
+
+export const mainConfig: Configuration = {
+  /**
+   * This is the main entry point for your application, it's the first file
+   * that runs in the main process.
+   */
+  entry: './src/index.ts',
+  // Put your normal webpack config below here
+  module: {
+    rules,
+  },
+  plugins,
+  resolve: {
+    extensions: ['.js', '.ts', '.jsx', '.tsx', '.css', '.json'],
+  },
+}

macapp/webpack.plugins.ts

@@ -0,0 +1,14 @@
+import type IForkTsCheckerWebpackPlugin from 'fork-ts-checker-webpack-plugin'
+import { DefinePlugin } from 'webpack'
+
+// eslint-disable-next-line @typescript-eslint/no-var-requires
+const ForkTsCheckerWebpackPlugin: typeof IForkTsCheckerWebpackPlugin = require('fork-ts-checker-webpack-plugin')
+
+export const plugins = [
+  new ForkTsCheckerWebpackPlugin({
+    logger: 'webpack-infrastructure',
+  }),
+  new DefinePlugin({
+    'process.env.TELEMETRY_WRITE_KEY': JSON.stringify(process.env.TELEMETRY_WRITE_KEY),
+  }),
+]

macapp/webpack.renderer.config.ts

@@ -0,0 +1,19 @@
+import type { Configuration } from 'webpack'
+
+import { rules } from './webpack.rules'
+import { plugins } from './webpack.plugins'
+
+rules.push({
+  test: /\.css$/,
+  use: [{ loader: 'style-loader' }, { loader: 'css-loader' }, { loader: 'postcss-loader' }],
+})
+
+export const rendererConfig: Configuration = {
+  module: {
+    rules,
+  },
+  plugins,
+  resolve: {
+    extensions: ['.js', '.ts', '.jsx', '.tsx', '.css'],
+  },
+}

macapp/webpack.rules.ts

@@ -0,0 +1,35 @@
+import type { ModuleOptions } from 'webpack'
+
+export const rules: Required<ModuleOptions>['rules'] = [
+  // Add support for native node modules
+  {
+    // We're specifying native_modules in the test because the asset relocator loader generates a
+    // "fake" .node file which is really a cjs file.
+    test: /native_modules[/\\].+\.node$/,
+    use: 'node-loader',
+  },
+  {
+    test: /[/\\]node_modules[/\\].+\.(m?js|node)$/,
+    parser: { amd: false },
+    use: {
+      loader: '@vercel/webpack-asset-relocator-loader',
+      options: {
+        outputAssetBase: 'native_modules',
+      },
+    },
+  },
+  {
+    test: /\.tsx?$/,
+    exclude: /(node_modules|\.webpack)/,
+    use: {
+      loader: 'ts-loader',
+      options: {
+        transpileOnly: true,
+      },
+    },
+  },
+  {
+    test: /\.svg$/,
+    use: ['@svgr/webpack'],
+  },
+]

parser/parser.go

@@ -6,8 +6,8 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"log/slog"
-	"slices"
+	"strconv"
+	"strings"
 )
 
 type Command struct {
@@ -15,118 +15,283 @@ type Command struct {
 	Args string
 }
 
-func (c *Command) Reset() {
-	c.Name = ""
-	c.Args = ""
-}
+type state int
 
-func Parse(reader io.Reader) ([]Command, error) {
-	var commands []Command
-	var command, modelCommand Command
+const (
+	stateNil state = iota
+	stateName
+	stateValue
+	stateParameter
+	stateMessage
+	stateComment
+)
 
-	scanner := bufio.NewScanner(reader)
-	scanner.Buffer(make([]byte, 0, bufio.MaxScanTokenSize), bufio.MaxScanTokenSize)
-	scanner.Split(scanModelfile)
-	for scanner.Scan() {
-		line := scanner.Bytes()
+var (
+	errMissingFrom        = errors.New("no FROM line")
+	errInvalidMessageRole = errors.New("message role must be one of \"system\", \"user\", or \"assistant\"")
+	errInvalidCommand     = errors.New("command must be one of \"from\", \"license\", \"template\", \"system\", \"adapter\", \"parameter\", or \"message\"")
+)
 
-		fields := bytes.SplitN(line, []byte(" "), 2)
-		if len(fields) == 0 || len(fields[0]) == 0 {
-			continue
-		}
+func Format(cmds []Command) string {
+	var sb strings.Builder
+	for _, cmd := range cmds {
+		name := cmd.Name
+		args := cmd.Args
 
-		switch string(bytes.ToUpper(fields[0])) {
-		case "FROM":
-			command.Name = "model"
-			command.Args = string(bytes.TrimSpace(fields[1]))
-			// copy command for validation
-			modelCommand = command
-		case "ADAPTER":
-			command.Name = string(bytes.ToLower(fields[0]))
-			command.Args = string(bytes.TrimSpace(fields[1]))
-		case "LICENSE", "TEMPLATE", "SYSTEM", "PROMPT":
-			command.Name = string(bytes.ToLower(fields[0]))
-			command.Args = string(fields[1])
-		case "PARAMETER":
-			fields = bytes.SplitN(fields[1], []byte(" "), 2)
-			if len(fields) < 2 {
-				return nil, fmt.Errorf("missing value for %s", fields)
-			}
-
-			command.Name = string(fields[0])
-			command.Args = string(bytes.TrimSpace(fields[1]))
-		case "EMBED":
-			return nil, fmt.Errorf("deprecated command: EMBED is no longer supported, use the /embed API endpoint instead")
-		case "MESSAGE":
-			command.Name = string(bytes.ToLower(fields[0]))
-			fields = bytes.SplitN(fields[1], []byte(" "), 2)
-			if len(fields) < 2 {
-				return nil, fmt.Errorf("should be in the format <role> <message>")
-			}
-			if !slices.Contains([]string{"system", "user", "assistant"}, string(bytes.ToLower(fields[0]))) {
-				return nil, fmt.Errorf("role must be one of \"system\", \"user\", or \"assistant\"")
-			}
-			command.Args = fmt.Sprintf("%s: %s", string(bytes.ToLower(fields[0])), string(fields[1]))
+		switch cmd.Name {
+		case "model":
+			name = "from"
+			args = cmd.Args
+		case "license", "template", "system", "adapter":
+			args = quote(args)
+		case "message":
+			role, message, _ := strings.Cut(cmd.Args, ": ")
+			args = role + " " + quote(message)
 		default:
-			if !bytes.HasPrefix(fields[0], []byte("#")) {
-				// log a warning for unknown commands
-				slog.Warn(fmt.Sprintf("Unknown command: %s", fields[0]))
-			}
-			continue
+			name = "parameter"
+			args = cmd.Name + " " + quote(cmd.Args)
 		}
 
-		commands = append(commands, command)
-		command.Reset()
+		fmt.Fprintln(&sb, strings.ToUpper(name), args)
 	}
 
-	if modelCommand.Args == "" {
-		return nil, errors.New("no FROM line for the model was specified")
-	}
-
-	return commands, scanner.Err()
+	return sb.String()
 }
 
-func scanModelfile(data []byte, atEOF bool) (advance int, token []byte, err error) {
-	advance, token, err = scan([]byte(`"""`), []byte(`"""`), data, atEOF)
-	if err != nil {
-		return 0, nil, err
-	}
+func Parse(r io.Reader) (cmds []Command, err error) {
+	var cmd Command
+	var curr state
+	var b bytes.Buffer
+	var role string
 
-	if advance > 0 && token != nil {
-		return advance, token, nil
-	}
-
-	advance, token, err = scan([]byte(`"`), []byte(`"`), data, atEOF)
-	if err != nil {
-		return 0, nil, err
-	}
-
-	if advance > 0 && token != nil {
-		return advance, token, nil
-	}
-
-	return bufio.ScanLines(data, atEOF)
-}
-
-func scan(openBytes, closeBytes, data []byte, atEOF bool) (advance int, token []byte, err error) {
-	newline := bytes.IndexByte(data, '\n')
-
-	if start := bytes.Index(data, openBytes); start >= 0 && start < newline {
-		end := bytes.Index(data[start+len(openBytes):], closeBytes)
-		if end < 0 {
-			if atEOF {
-				return 0, nil, fmt.Errorf("unterminated %s: expecting %s", openBytes, closeBytes)
-			} else {
-				return 0, nil, nil
-			}
+	br := bufio.NewReader(r)
+	for {
+		r, _, err := br.ReadRune()
+		if errors.Is(err, io.EOF) {
+			break
+		} else if err != nil {
+			return nil, err
 		}
 
-		n := start + len(openBytes) + end + len(closeBytes)
+		next, r, err := parseRuneForState(r, curr)
+		if errors.Is(err, io.ErrUnexpectedEOF) {
+			return nil, fmt.Errorf("%w: %s", err, b.String())
+		} else if err != nil {
+			return nil, err
+		}
 
-		newData := data[:start]
-		newData = append(newData, data[start+len(openBytes):n-len(closeBytes)]...)
-		return n, newData, nil
+		// process the state transition, some transitions need to be intercepted and redirected
+		if next != curr {
+			switch curr {
+			case stateName:
+				if !isValidCommand(b.String()) {
+					return nil, errInvalidCommand
+				}
+
+				// next state sometimes depends on the current buffer value
+				switch s := strings.ToLower(b.String()); s {
+				case "from":
+					cmd.Name = "model"
+				case "parameter":
+					// transition to stateParameter which sets command name
+					next = stateParameter
+				case "message":
+					// transition to stateMessage which validates the message role
+					next = stateMessage
+					fallthrough
+				default:
+					cmd.Name = s
+				}
+			case stateParameter:
+				cmd.Name = b.String()
+			case stateMessage:
+				if !isValidMessageRole(b.String()) {
+					return nil, errInvalidMessageRole
+				}
+
+				role = b.String()
+			case stateComment, stateNil:
+				// pass
+			case stateValue:
+				s, ok := unquote(b.String())
+				if !ok || isSpace(r) {
+					if _, err := b.WriteRune(r); err != nil {
+						return nil, err
+					}
+
+					continue
+				}
+
+				if role != "" {
+					s = role + ": " + s
+					role = ""
+				}
+
+				cmd.Args = s
+				cmds = append(cmds, cmd)
+			}
+
+			b.Reset()
+			curr = next
+		}
+
+		if strconv.IsPrint(r) {
+			if _, err := b.WriteRune(r); err != nil {
+				return nil, err
+			}
+		}
 	}
 
-	return 0, nil, nil
+	// flush the buffer
+	switch curr {
+	case stateComment, stateNil:
+		// pass; nothing to flush
+	case stateValue:
+		s, ok := unquote(b.String())
+		if !ok {
+			return nil, io.ErrUnexpectedEOF
+		}
+
+		if role != "" {
+			s = role + ": " + s
+		}
+
+		cmd.Args = s
+		cmds = append(cmds, cmd)
+	default:
+		return nil, io.ErrUnexpectedEOF
+	}
+
+	for _, cmd := range cmds {
+		if cmd.Name == "model" {
+			return cmds, nil
+		}
+	}
+
+	return nil, errMissingFrom
+}
+
+func parseRuneForState(r rune, cs state) (state, rune, error) {
+	switch cs {
+	case stateNil:
+		switch {
+		case r == '#':
+			return stateComment, 0, nil
+		case isSpace(r), isNewline(r):
+			return stateNil, 0, nil
+		default:
+			return stateName, r, nil
+		}
+	case stateName:
+		switch {
+		case isAlpha(r):
+			return stateName, r, nil
+		case isSpace(r):
+			return stateValue, 0, nil
+		default:
+			return stateNil, 0, errInvalidCommand
+		}
+	case stateValue:
+		switch {
+		case isNewline(r):
+			return stateNil, r, nil
+		case isSpace(r):
+			return stateNil, r, nil
+		default:
+			return stateValue, r, nil
+		}
+	case stateParameter:
+		switch {
+		case isAlpha(r), isNumber(r), r == '_':
+			return stateParameter, r, nil
+		case isSpace(r):
+			return stateValue, 0, nil
+		default:
+			return stateNil, 0, io.ErrUnexpectedEOF
+		}
+	case stateMessage:
+		switch {
+		case isAlpha(r):
+			return stateMessage, r, nil
+		case isSpace(r):
+			return stateValue, 0, nil
+		default:
+			return stateNil, 0, io.ErrUnexpectedEOF
+		}
+	case stateComment:
+		switch {
+		case isNewline(r):
+			return stateNil, 0, nil
+		default:
+			return stateComment, 0, nil
+		}
+	default:
+		return stateNil, 0, errors.New("")
+	}
+}
+
+func quote(s string) string {
+	if strings.Contains(s, "\n") || strings.HasPrefix(s, " ") || strings.HasSuffix(s, " ") {
+		if strings.Contains(s, "\"") {
+			return `"""` + s + `"""`
+		}
+
+		return `"` + s + `"`
+	}
+
+	return s
+}
+
+func unquote(s string) (string, bool) {
+	if len(s) == 0 {
+		return "", false
+	}
+
+	// TODO: single quotes
+	if len(s) >= 3 && s[:3] == `"""` {
+		if len(s) >= 6 && s[len(s)-3:] == `"""` {
+			return s[3 : len(s)-3], true
+		}
+
+		return "", false
+	}
+
+	if len(s) >= 1 && s[0] == '"' {
+		if len(s) >= 2 && s[len(s)-1] == '"' {
+			return s[1 : len(s)-1], true
+		}
+
+		return "", false
+	}
+
+	return s, true
+}
+
+func isAlpha(r rune) bool {
+	return r >= 'a' && r <= 'z' || r >= 'A' && r <= 'Z'
+}
+
+func isNumber(r rune) bool {
+	return r >= '0' && r <= '9'
+}
+
+func isSpace(r rune) bool {
+	return r == ' ' || r == '\t'
+}
+
+func isNewline(r rune) bool {
+	return r == '\r' || r == '\n'
+}
+
+func isValidMessageRole(role string) bool {
+	return role == "system" || role == "user" || role == "assistant"
+}
+
+func isValidCommand(cmd string) bool {
+	switch strings.ToLower(cmd) {
+	case "from", "license", "template", "system", "adapter", "parameter", "message":
+		return true
+	default:
+		return false
+	}
 }

parser/parser_test.go

@@ -1,14 +1,16 @@
 package parser
 
 import (
+	"bytes"
+	"fmt"
+	"io"
 	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
 )
 
-func Test_Parser(t *testing.T) {
-
+func TestParser(t *testing.T) {
 	input := `
 FROM model1
 ADAPTER adapter1
@@ -35,21 +37,62 @@ TEMPLATE template1
 	assert.Equal(t, expectedCommands, commands)
 }
 
-func Test_Parser_NoFromLine(t *testing.T) {
+func TestParserFrom(t *testing.T) {
+	var cases = []struct {
+		input    string
+		expected []Command
+		err      error
+	}{
+		{
+			"FROM foo",
+			[]Command{{Name: "model", Args: "foo"}},
+			nil,
+		},
+		{
+			"FROM /path/to/model",
+			[]Command{{Name: "model", Args: "/path/to/model"}},
+			nil,
+		},
+		{
+			"FROM /path/to/model/fp16.bin",
+			[]Command{{Name: "model", Args: "/path/to/model/fp16.bin"}},
+			nil,
+		},
+		{
+			"FROM llama3:latest",
+			[]Command{{Name: "model", Args: "llama3:latest"}},
+			nil,
+		},
+		{
+			"FROM llama3:7b-instruct-q4_K_M",
+			[]Command{{Name: "model", Args: "llama3:7b-instruct-q4_K_M"}},
+			nil,
+		},
+		{
+			"", nil, errMissingFrom,
+		},
+		{
+			"PARAMETER param1 value1",
+			nil,
+			errMissingFrom,
+		},
+		{
+			"PARAMETER param1 value1\nFROM foo",
+			[]Command{{Name: "param1", Args: "value1"}, {Name: "model", Args: "foo"}},
+			nil,
+		},
+	}
 
-	input := `
-PARAMETER param1 value1
-PARAMETER param2 value2
-`
-
-	reader := strings.NewReader(input)
-
-	_, err := Parse(reader)
-	assert.ErrorContains(t, err, "no FROM line")
+	for _, c := range cases {
+		t.Run("", func(t *testing.T) {
+			commands, err := Parse(strings.NewReader(c.input))
+			assert.ErrorIs(t, err, c.err)
+			assert.Equal(t, c.expected, commands)
+		})
+	}
 }
 
-func Test_Parser_MissingValue(t *testing.T) {
-
+func TestParserParametersMissingValue(t *testing.T) {
 	input := `
 FROM foo
 PARAMETER param1
@@ -58,41 +101,401 @@ PARAMETER param1
 	reader := strings.NewReader(input)
 
 	_, err := Parse(reader)
-	assert.ErrorContains(t, err, "missing value for [param1]")
+	assert.ErrorIs(t, err, io.ErrUnexpectedEOF)
+}
+
+func TestParserBadCommand(t *testing.T) {
+	input := `
+FROM foo
+BADCOMMAND param1 value1
+`
+	_, err := Parse(strings.NewReader(input))
+	assert.ErrorIs(t, err, errInvalidCommand)
 
 }
 
-func Test_Parser_Messages(t *testing.T) {
-
-	input := `
+func TestParserMessages(t *testing.T) {
+	var cases = []struct {
+		input    string
+		expected []Command
+		err      error
+	}{
+		{
+			`
+FROM foo
+MESSAGE system You are a Parser. Always Parse things.
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "message", Args: "system: You are a Parser. Always Parse things."},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+MESSAGE system You are a Parser. Always Parse things.`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "message", Args: "system: You are a Parser. Always Parse things."},
+			},
+			nil,
+		},
+		{
+			`
 FROM foo
 MESSAGE system You are a Parser. Always Parse things.
 MESSAGE user Hey there!
 MESSAGE assistant Hello, I want to parse all the things!
-`
-
-	reader := strings.NewReader(input)
-	commands, err := Parse(reader)
-	assert.Nil(t, err)
-
-	expectedCommands := []Command{
-		{Name: "model", Args: "foo"},
-		{Name: "message", Args: "system: You are a Parser. Always Parse things."},
-		{Name: "message", Args: "user: Hey there!"},
-		{Name: "message", Args: "assistant: Hello, I want to parse all the things!"},
-	}
-
-	assert.Equal(t, expectedCommands, commands)
-}
-
-func Test_Parser_Messages_BadRole(t *testing.T) {
-
-	input := `
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "message", Args: "system: You are a Parser. Always Parse things."},
+				{Name: "message", Args: "user: Hey there!"},
+				{Name: "message", Args: "assistant: Hello, I want to parse all the things!"},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+MESSAGE system """
+You are a multiline Parser. Always Parse things.
+"""
+			`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "message", Args: "system: \nYou are a multiline Parser. Always Parse things.\n"},
+			},
+			nil,
+		},
+		{
+			`
 FROM foo
 MESSAGE badguy I'm a bad guy!
-`
+`,
+			nil,
+			errInvalidMessageRole,
+		},
+		{
+			`
+FROM foo
+MESSAGE system
+`,
+			nil,
+			io.ErrUnexpectedEOF,
+		},
+		{
+			`
+FROM foo
+MESSAGE system`,
+			nil,
+			io.ErrUnexpectedEOF,
+		},
+	}
+
+	for _, c := range cases {
+		t.Run("", func(t *testing.T) {
+			commands, err := Parse(strings.NewReader(c.input))
+			assert.ErrorIs(t, err, c.err)
+			assert.Equal(t, c.expected, commands)
+		})
+	}
+}
+
+func TestParserQuoted(t *testing.T) {
+	var cases = []struct {
+		multiline string
+		expected  []Command
+		err       error
+	}{
+		{
+			`
+FROM foo
+SYSTEM """
+This is a
+multiline system.
+"""
+			`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "\nThis is a\nmultiline system.\n"},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """
+This is a
+multiline system."""
+			`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "\nThis is a\nmultiline system."},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """This is a
+multiline system."""
+			`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "This is a\nmultiline system."},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """This is a multiline system."""
+			`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "This is a multiline system."},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """This is a multiline system.""
+			`,
+			nil,
+			io.ErrUnexpectedEOF,
+		},
+		{
+			`
+FROM foo
+SYSTEM "
+			`,
+			nil,
+			io.ErrUnexpectedEOF,
+		},
+		{
+			`
+FROM foo
+SYSTEM """
+This is a multiline system with "quotes".
+"""
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "\nThis is a multiline system with \"quotes\".\n"},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """"""
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: ""},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM ""
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: ""},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM "'"
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: "'"},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+SYSTEM """''"'""'""'"'''''""'""'"""
+`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "system", Args: `''"'""'""'"'''''""'""'`},
+			},
+			nil,
+		},
+		{
+			`
+FROM foo
+TEMPLATE """
+{{ .Prompt }}
+"""`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+				{Name: "template", Args: "\n{{ .Prompt }}\n"},
+			},
+			nil,
+		},
+	}
+
+	for _, c := range cases {
+		t.Run("", func(t *testing.T) {
+			commands, err := Parse(strings.NewReader(c.multiline))
+			assert.ErrorIs(t, err, c.err)
+			assert.Equal(t, c.expected, commands)
+		})
+	}
+}
+
+func TestParserParameters(t *testing.T) {
+	var cases = map[string]struct {
+		name, value string
+	}{
+		"numa true":                    {"numa", "true"},
+		"num_ctx 1":                    {"num_ctx", "1"},
+		"num_batch 1":                  {"num_batch", "1"},
+		"num_gqa 1":                    {"num_gqa", "1"},
+		"num_gpu 1":                    {"num_gpu", "1"},
+		"main_gpu 1":                   {"main_gpu", "1"},
+		"low_vram true":                {"low_vram", "true"},
+		"f16_kv true":                  {"f16_kv", "true"},
+		"logits_all true":              {"logits_all", "true"},
+		"vocab_only true":              {"vocab_only", "true"},
+		"use_mmap true":                {"use_mmap", "true"},
+		"use_mlock true":               {"use_mlock", "true"},
+		"num_thread 1":                 {"num_thread", "1"},
+		"num_keep 1":                   {"num_keep", "1"},
+		"seed 1":                       {"seed", "1"},
+		"num_predict 1":                {"num_predict", "1"},
+		"top_k 1":                      {"top_k", "1"},
+		"top_p 1.0":                    {"top_p", "1.0"},
+		"tfs_z 1.0":                    {"tfs_z", "1.0"},
+		"typical_p 1.0":                {"typical_p", "1.0"},
+		"repeat_last_n 1":              {"repeat_last_n", "1"},
+		"temperature 1.0":              {"temperature", "1.0"},
+		"repeat_penalty 1.0":           {"repeat_penalty", "1.0"},
+		"presence_penalty 1.0":         {"presence_penalty", "1.0"},
+		"frequency_penalty 1.0":        {"frequency_penalty", "1.0"},
+		"mirostat 1":                   {"mirostat", "1"},
+		"mirostat_tau 1.0":             {"mirostat_tau", "1.0"},
+		"mirostat_eta 1.0":             {"mirostat_eta", "1.0"},
+		"penalize_newline true":        {"penalize_newline", "true"},
+		"stop ### User:":               {"stop", "### User:"},
+		"stop ### User: ":              {"stop", "### User: "},
+		"stop \"### User:\"":           {"stop", "### User:"},
+		"stop \"### User: \"":          {"stop", "### User: "},
+		"stop \"\"\"### User:\"\"\"":   {"stop", "### User:"},
+		"stop \"\"\"### User:\n\"\"\"": {"stop", "### User:\n"},
+		"stop <|endoftext|>":           {"stop", "<|endoftext|>"},
+		"stop <|eot_id|>":              {"stop", "<|eot_id|>"},
+		"stop </s>":                    {"stop", "</s>"},
+	}
+
+	for k, v := range cases {
+		t.Run(k, func(t *testing.T) {
+			var b bytes.Buffer
+			fmt.Fprintln(&b, "FROM foo")
+			fmt.Fprintln(&b, "PARAMETER", k)
+			commands, err := Parse(&b)
+			assert.Nil(t, err)
+
+			assert.Equal(t, []Command{
+				{Name: "model", Args: "foo"},
+				{Name: v.name, Args: v.value},
+			}, commands)
+		})
+	}
+}
+
+func TestParserComments(t *testing.T) {
+	var cases = []struct {
+		input    string
+		expected []Command
+	}{
+		{
+			`
+# comment
+FROM foo
+	`,
+			[]Command{
+				{Name: "model", Args: "foo"},
+			},
+		},
+	}
+
+	for _, c := range cases {
+		t.Run("", func(t *testing.T) {
+			commands, err := Parse(strings.NewReader(c.input))
+			assert.Nil(t, err)
+			assert.Equal(t, c.expected, commands)
+		})
+	}
+}
+
+func TestParseFormatParse(t *testing.T) {
+	var cases = []string{
+		`
+FROM foo
+ADAPTER adapter1
+LICENSE MIT
+PARAMETER param1 value1
+PARAMETER param2 value2
+TEMPLATE template1
+MESSAGE system You are a Parser. Always Parse things.
+MESSAGE user Hey there!
+MESSAGE assistant Hello, I want to parse all the things!
+`,
+		`
+FROM foo
+ADAPTER adapter1
+LICENSE MIT
+PARAMETER param1 value1
+PARAMETER param2 value2
+TEMPLATE template1
+MESSAGE system """
+You are a store greeter. Always responsed with "Hello!".
+"""
+MESSAGE user Hey there!
+MESSAGE assistant Hello, I want to parse all the things!
+`,
+		`
+FROM foo
+ADAPTER adapter1
+LICENSE """
+Very long and boring legal text.
+Blah blah blah.
+"Oh look, a quote!"
+"""
+
+PARAMETER param1 value1
+PARAMETER param2 value2
+TEMPLATE template1
+MESSAGE system """
+You are a store greeter. Always responsed with "Hello!".
+"""
+MESSAGE user Hey there!
+MESSAGE assistant Hello, I want to parse all the things!
+`,
+	}
+
+	for _, c := range cases {
+		t.Run("", func(t *testing.T) {
+			commands, err := Parse(strings.NewReader(c))
+			assert.NoError(t, err)
+
+			commands2, err := Parse(strings.NewReader(Format(commands)))
+			assert.NoError(t, err)
+
+			assert.Equal(t, commands, commands2)
+		})
+	}
 
-	reader := strings.NewReader(input)
-	_, err := Parse(reader)
-	assert.ErrorContains(t, err, "role must be one of \"system\", \"user\", or \"assistant\"")
 }

scripts/build_darwin.sh

@@ -11,37 +11,26 @@ for TARGETARCH in arm64 amd64; do
     rm -rf llm/llama.cpp/build
     GOOS=darwin GOARCH=$TARGETARCH go generate ./...
     CGO_ENABLED=1 GOOS=darwin GOARCH=$TARGETARCH go build -trimpath -o dist/ollama-darwin-$TARGETARCH
-    CGO_ENABLED=1 GOOS=darwin GOARCH=$TARGETARCH go build -C app -trimpath -o ../dist/ollama-app-darwin-$TARGETARCH
+    CGO_ENABLED=1 GOOS=darwin GOARCH=$TARGETARCH go build -trimpath -cover -o dist/ollama-darwin-$TARGETARCH-cov
 done
 
 lipo -create -output dist/ollama dist/ollama-darwin-arm64 dist/ollama-darwin-amd64
-lipo -create -output dist/ollama-app dist/ollama-app-darwin-arm64 dist/ollama-app-darwin-amd64
-rm -f dist/ollama-darwin-* dist/ollama-app-darwin-*
-
-# create the mac app
-rm -rf dist/Ollama.app
-cp -R app/darwin/Ollama.app dist/
-/usr/libexec/PlistBuddy -c "Set :CFBundleShortVersionString $VERSION" dist/Ollama.app/Contents/Info.plist
-mkdir -p dist/Ollama.app/Contents/MacOS
-mv dist/ollama-app dist/Ollama.app/Contents/MacOS/Ollama
-cp dist/ollama dist/Ollama.app/Contents/Resources/ollama
-
-# sign and notarize the app
+rm -f dist/ollama-darwin-arm64 dist/ollama-darwin-amd64
 if [ -n "$APPLE_IDENTITY" ]; then
-    codesign -f --timestamp --options=runtime --sign "$APPLE_IDENTITY" --identifier ai.ollama.ollama dist/Ollama.app/Contents/MacOS/Ollama
-    codesign -f --timestamp --options=runtime --sign "$APPLE_IDENTITY" --identifier ai.ollama.ollama dist/Ollama.app/Contents/Resources/ollama
-    codesign -f --timestamp --options=runtime --sign "$APPLE_IDENTITY" --identifier ai.ollama.ollama dist/Ollama.app
-    ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
-    rm -rf dist/Ollama.app
-    xcrun notarytool submit dist/Ollama-darwin.zip --wait --timeout 10m --apple-id $APPLE_ID --password $APPLE_PASSWORD --team-id $APPLE_TEAM_ID
-    unzip dist/Ollama-darwin.zip -d dist
-    rm -f dist/Ollama-darwin.zip
-    xcrun stapler staple "dist/Ollama.app"
-    ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
-    rm -rf dist/Ollama.app
+    codesign --deep --force --options=runtime --sign "$APPLE_IDENTITY" --timestamp dist/ollama
 else
     echo "Skipping code signing - set APPLE_IDENTITY"
 fi
+chmod +x dist/ollama
+
+# build and optionally sign the mac app
+npm install --prefix macapp
+if [ -n "$APPLE_IDENTITY" ]; then
+    npm run --prefix macapp make:sign
+else 
+    npm run --prefix macapp make
+fi
+cp macapp/out/make/zip/darwin/universal/Ollama-darwin-universal-$VERSION.zip dist/Ollama-darwin.zip
 
 # sign the binary and rename it
 if [ -n "$APPLE_IDENTITY" ]; then

scripts/build_windows.ps1

@@ -7,6 +7,8 @@
 $ErrorActionPreference = "Stop"
 
 function checkEnv() {
+    $script:TARGET_ARCH=$Env:PROCESSOR_ARCHITECTURE.ToLower()
+    Write-host "Building for ${script:TARGET_ARCH}"
     write-host "Locating required tools and paths"
     $script:SRC_DIR=$PWD
     if (!$env:VCToolsRedistDir) {
@@ -27,10 +29,10 @@ function checkEnv() {
     } else {
         $script:NVIDIA_DIR=$env:NVIDIA_DIR
     }
-
+    
     $script:INNO_SETUP_DIR=(get-item "C:\Program Files*\Inno Setup*\")[0]
 
-    $script:DEPS_DIR="${script:SRC_DIR}\dist\windows-amd64"
+    $script:DEPS_DIR="${script:SRC_DIR}\dist\windows-${script:TARGET_ARCH}"
     $env:CGO_ENABLED="1"
     echo "Checking version"
     if (!$env:VERSION) {
@@ -70,7 +72,7 @@ function buildOllama() {
     write-host "Building ollama CLI"
     if ($null -eq ${env:OLLAMA_SKIP_GENERATE}) {
         & go generate ./...
-        if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
+        if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}    
     } else {
         write-host "Skipping generate step with OLLAMA_SKIP_GENERATE set"
     }
@@ -81,15 +83,15 @@ function buildOllama() {
             /csp "Google Cloud KMS Provider" /kc ${env:KEY_CONTAINER} ollama.exe
         if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
     }
-    New-Item -ItemType Directory -Path .\dist\windows-amd64\ -Force
-    cp .\ollama.exe .\dist\windows-amd64\
+    New-Item -ItemType Directory -Path .\dist\windows-${script:TARGET_ARCH}\ -Force
+    cp .\ollama.exe .\dist\windows-${script:TARGET_ARCH}\
 }
 
 function buildApp() {
     write-host "Building Ollama App"
     cd "${script:SRC_DIR}\app"
-    & windres -l 0 -o ollama.syso windows\ollama.rc
-    & go build -trimpath -ldflags "-s -w -H windowsgui -X=github.com/jmorganca/ollama/version.Version=$script:VERSION -X=github.com/jmorganca/ollama/server.mode=release" .
+    & windres -l 0 -o ollama.syso ollama.rc
+    & go build -trimpath -ldflags "-s -w -H windowsgui -X=github.com/ollama/ollama/version.Version=$script:VERSION -X=github.com/ollama/ollama/server.mode=release" .
     if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
     if ("${env:KEY_CONTAINER}") {
         & "${script:SignTool}" sign /v /fd sha256 /t http://timestamp.digicert.com /f "${script:OLLAMA_CERT}" `
@@ -110,7 +112,7 @@ function gatherDependencies() {
     cp "${env:VCToolsRedistDir}\x64\Microsoft.VC*.CRT\vcruntime140_1.dll" "${script:DEPS_DIR}\"
 
 
-    cp "${script:SRC_DIR}\app\windows\ollama_welcome.ps1" "${script:SRC_DIR}\dist\"
+    cp "${script:SRC_DIR}\app\ollama_welcome.ps1" "${script:SRC_DIR}\dist\"
     if ("${env:KEY_CONTAINER}") {
         write-host "about to sign"
         foreach ($file in (get-childitem "${script:DEPS_DIR}/cu*.dll") + @("${script:SRC_DIR}\dist\ollama_welcome.ps1")){
@@ -127,16 +129,16 @@ function buildInstaller() {
     cd "${script:SRC_DIR}\app"
     $env:PKG_VERSION=$script:PKG_VERSION
     if ("${env:KEY_CONTAINER}") {
-        & "${script:INNO_SETUP_DIR}\ISCC.exe" /SMySignTool="${script:SignTool} sign /fd sha256 /t http://timestamp.digicert.com /f ${script:OLLAMA_CERT} /csp `$qGoogle Cloud KMS Provider`$q /kc ${env:KEY_CONTAINER} `$f" .\windows\ollama.iss
+        & "${script:INNO_SETUP_DIR}\ISCC.exe" /DARCH=$script:TARGET_ARCH /SMySignTool="${script:SignTool} sign /fd sha256 /t http://timestamp.digicert.com /f ${script:OLLAMA_CERT} /csp `$qGoogle Cloud KMS Provider`$q /kc ${env:KEY_CONTAINER} `$f" .\ollama.iss
     } else {
-        & "${script:INNO_SETUP_DIR}\ISCC.exe" .\windows\ollama.iss
+        & "${script:INNO_SETUP_DIR}\ISCC.exe" /DARCH=$script:TARGET_ARCH .\ollama.iss
     }
     if ($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
+    write-host "Generating stand-alone distribution zip file ${script:SRC_DIR}\dist\ollama-windows-${script:TARGET_ARCH}.zip"
+    Compress-Archive -Path "${script:SRC_DIR}\dist\windows-${script:TARGET_ARCH}\*" -DestinationPath "${script:SRC_DIR}\dist\ollama-windows-${script:TARGET_ARCH}.zip" -Force
 }
 
 try {

scripts/publish.sh

@@ -0,0 +1,25 @@
+# Set your variables here.
+REPO="jmorganca/ollama"
+
+# Check if VERSION is set
+if [[ -z "${VERSION}" ]]; then
+  echo "VERSION is not set. Please set the VERSION environment variable."
+  exit 1
+fi
+
+OS=$(go env GOOS)
+
+./script/build_${OS}.sh
+
+# Create a new tag if it doesn't exist.
+if ! git rev-parse v$VERSION >/dev/null 2>&1; then
+  git tag v$VERSION
+fi
+
+git push origin v$VERSION
+
+# Create a new release.
+gh release create -p v$VERSION -t v$VERSION
+
+# Upload the zip file.
+gh release upload v$VERSION ./dist/* --clobber

scripts/run_darwin.sh

@@ -1,10 +0,0 @@
-#!/bin/bash
-
-set -e
-
-rm -rf $TMPDIR/Ollama.app
-cp -R app/darwin/Ollama.app $TMPDIR/Ollama.app
-mkdir -p $TMPDIR/Ollama.app/Contents/Resources $TMPDIR/Ollama.app/Contents/MacOS
-go build -o $TMPDIR/Ollama.app/Contents/Resources/ollama .
-go build -C app -o $TMPDIR/Ollama.app/Contents/MacOS/Ollama .
-$TMPDIR/Ollama.app/Contents/MacOS/Ollama

server/layer.go

@@ -5,39 +5,18 @@ import (
 	"fmt"
 	"io"
 	"os"
-	"strings"
-
-	"golang.org/x/exp/slices"
 )
 
-type Layers struct {
-	items []*Layer
-}
-
-func (ls *Layers) Add(layer *Layer) {
-	if layer.Size > 0 {
-		ls.items = append(ls.items, layer)
-	}
-}
-
-func (ls *Layers) Replace(layer *Layer) {
-	if layer.Size > 0 {
-		mediatype := layer.MediaType
-		layers := slices.DeleteFunc(ls.items, func(l *Layer) bool {
-			return l.MediaType == mediatype
-		})
-
-		ls.items = append(layers, layer)
-	}
-}
-
 type Layer struct {
 	MediaType string `json:"mediaType"`
 	Digest    string `json:"digest"`
 	Size      int64  `json:"size"`
 	From      string `json:"from,omitempty"`
 
-	tempFileName string
+	Intermediate bool   `json:"intermediate,omitempty"`
+	MergeBase    string `json:"merge_base,omitempty"`
+
+	message string
 }
 
 func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
@@ -46,14 +25,12 @@ func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
 		return nil, err
 	}
 
-	const delimiter = "-"
-
-	pattern := strings.Join([]string{"sha256", "*-partial"}, delimiter)
-	temp, err := os.CreateTemp(blobs, pattern)
+	temp, err := os.CreateTemp(blobs, "sha256-")
 	if err != nil {
 		return nil, err
 	}
 	defer temp.Close()
+	defer os.Remove(temp.Name())
 
 	sha256sum := sha256.New()
 	n, err := io.Copy(io.MultiWriter(temp, sha256sum), r)
@@ -61,11 +38,29 @@ func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
 		return nil, err
 	}
 
+	if err := temp.Close(); err != nil {
+		return nil, err
+	}
+
+	digest := fmt.Sprintf("sha256:%x", sha256sum.Sum(nil))
+	blob, err := GetBlobsPath(digest)
+	if err != nil {
+		return nil, err
+	}
+
+	status := "using existing layer"
+	if _, err := os.Stat(blob); err != nil {
+		status = "creating new layer"
+		if err := os.Rename(temp.Name(), blob); err != nil {
+			return nil, err
+		}
+	}
+
 	return &Layer{
-		MediaType:    mediatype,
-		Digest:       fmt.Sprintf("sha256:%x", sha256sum.Sum(nil)),
-		Size:         n,
-		tempFileName: temp.Name(),
+		MediaType: mediatype,
+		Digest:    digest,
+		Size:      n,
+		message:   fmt.Sprintf("%s %s", status, digest),
 	}, nil
 }
 
@@ -85,21 +80,15 @@ func NewLayerFromLayer(digest, mediatype, from string) (*Layer, error) {
 		Digest:    digest,
 		Size:      fi.Size(),
 		From:      from,
+		message:   fmt.Sprintf("using existing layer %s", digest),
 	}, nil
 }
 
-func (l *Layer) Commit() (bool, error) {
-	// always remove temp
-	defer os.Remove(l.tempFileName)
-
+func (l *Layer) Open() (*os.File, error) {
 	blob, err := GetBlobsPath(l.Digest)
 	if err != nil {
-		return false, err
+		return nil, err
 	}
 
-	if _, err := os.Stat(blob); err != nil {
-		return true, os.Rename(l.tempFileName, blob)
-	}
-
-	return false, nil
+	return os.Open(blob)
 }

server/model.go

@@ -0,0 +1,259 @@
+package server
+
+import (
+	"archive/zip"
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+
+	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/convert"
+	"github.com/ollama/ollama/llm"
+	"github.com/ollama/ollama/types/model"
+)
+
+type layerWithGGML struct {
+	*Layer
+	*llm.GGML
+}
+
+func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
+	modelpath := ParseModelPath(name.String())
+	manifest, _, err := GetManifest(modelpath)
+	switch {
+	case errors.Is(err, os.ErrNotExist):
+		if err := PullModel(ctx, name.String(), &registryOptions{}, fn); err != nil {
+			return nil, err
+		}
+
+		modelpath = ParseModelPath(name.String())
+		manifest, _, err = GetManifest(modelpath)
+		if err != nil {
+			return nil, err
+		}
+	case err != nil:
+		return nil, err
+	}
+
+	for _, layer := range manifest.Layers {
+		layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, modelpath.GetShortTagname())
+		if err != nil {
+			return nil, err
+		}
+
+		switch layer.MediaType {
+		case "application/vnd.ollama.image.model",
+			"application/vnd.ollama.image.projector",
+			"application/vnd.ollama.image.adapter":
+			blobpath, err := GetBlobsPath(layer.Digest)
+			if err != nil {
+				return nil, err
+			}
+
+			blob, err := os.Open(blobpath)
+			if err != nil {
+				return nil, err
+			}
+			defer blob.Close()
+
+			ggml, _, err := llm.DecodeGGML(blob)
+			if err != nil {
+				return nil, err
+			}
+
+			layers = append(layers, &layerWithGGML{layer, ggml})
+		default:
+			layers = append(layers, &layerWithGGML{layer, nil})
+		}
+
+	}
+
+	return layers, nil
+}
+
+func parseFromZipFile(_ context.Context, file *os.File, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
+	stat, err := file.Stat()
+	if err != nil {
+		return nil, err
+	}
+
+	r, err := zip.NewReader(file, stat.Size())
+	if err != nil {
+		return nil, err
+	}
+
+	tempdir, err := os.MkdirTemp(filepath.Dir(file.Name()), "")
+	if err != nil {
+		return nil, err
+	}
+	defer os.RemoveAll(tempdir)
+
+	fn(api.ProgressResponse{Status: "unpacking model metadata"})
+	for _, f := range r.File {
+		// TODO(mxyng): this should not write out all files to disk
+		outfile, err := os.Create(filepath.Join(tempdir, f.Name))
+		if err != nil {
+			return nil, err
+		}
+
+		infile, err := f.Open()
+		if err != nil {
+			return nil, err
+		}
+
+		if _, err = io.Copy(outfile, infile); err != nil {
+			return nil, err
+		}
+
+		if err := outfile.Close(); err != nil {
+			return nil, err
+		}
+
+		if err := infile.Close(); err != nil {
+			return nil, err
+		}
+	}
+
+	mf, err := convert.GetModelFormat(tempdir)
+	if err != nil {
+		return nil, err
+	}
+
+	params, err := mf.GetParams(tempdir)
+	if err != nil {
+		return nil, err
+	}
+
+	mArch, err := mf.GetModelArch("", tempdir, params)
+	if err != nil {
+		return nil, err
+	}
+
+	fn(api.ProgressResponse{Status: "processing tensors"})
+	if err := mArch.GetTensors(); err != nil {
+		return nil, err
+	}
+
+	if err := mArch.LoadVocab(); err != nil {
+		return nil, err
+	}
+
+	fn(api.ProgressResponse{Status: "converting model"})
+
+	// TODO(mxyng): this should write directly into a layer
+	// e.g. NewLayer(arch.Reader(), "application/vnd.ollama.image.model")
+	temp, err := os.CreateTemp(tempdir, "fp16")
+	if err != nil {
+		return nil, err
+	}
+	defer temp.Close()
+	defer os.Remove(temp.Name())
+
+	if err = mArch.WriteGGUF(temp); err != nil {
+		return nil, err
+	}
+
+	if _, err := temp.Seek(0, io.SeekStart); err != nil {
+		return nil, err
+	}
+
+	layer, err := NewLayer(temp, "application/vnd.ollama.image.model")
+	if err != nil {
+		return nil, fmt.Errorf("aaa: %w", err)
+	}
+
+	blobpath, err := GetBlobsPath(layer.Digest)
+	if err != nil {
+		return nil, err
+	}
+
+	bin, err := os.Open(blobpath)
+	if err != nil {
+		return nil, err
+	}
+	defer bin.Close()
+
+	ggml, _, err := llm.DecodeGGML(bin)
+	if err != nil {
+		return nil, err
+	}
+
+	layer, err = NewLayerFromLayer(layer.Digest, layer.MediaType, "")
+	if err != nil {
+		return nil, err
+	}
+
+	layers = append(layers, &layerWithGGML{layer, ggml})
+	return layers, nil
+}
+
+func parseFromFile(ctx context.Context, file *os.File, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
+	sr := io.NewSectionReader(file, 0, 512)
+	contentType, err := detectContentType(sr)
+	if err != nil {
+		return nil, err
+	}
+
+	switch contentType {
+	case "gguf", "ggla":
+		// noop
+	case "application/zip":
+		return parseFromZipFile(ctx, file, fn)
+	default:
+		return nil, fmt.Errorf("unsupported content type: %s", contentType)
+	}
+
+	stat, err := file.Stat()
+	if err != nil {
+		return nil, err
+	}
+
+	var offset int64
+	for offset < stat.Size() {
+		ggml, n, err := llm.DecodeGGML(file)
+		if errors.Is(err, io.EOF) {
+			break
+		} else if err != nil {
+			return nil, err
+		}
+
+		mediatype := "application/vnd.ollama.image.model"
+		if ggml.Name() == "ggla" {
+			mediatype = "application/vnd.ollama.image.adapter"
+		} else if ggml.KV().Architecture() == "clip" {
+			mediatype = "application/vnd.ollama.image.projector"
+		}
+
+		layer, err := NewLayer(io.NewSectionReader(file, offset, n), mediatype)
+		if err != nil {
+			return nil, err
+		}
+
+		layers = append(layers, &layerWithGGML{layer, ggml})
+		offset = n
+	}
+
+	return layers, nil
+}
+
+func detectContentType(r io.Reader) (string, error) {
+	var b bytes.Buffer
+	if _, err := io.Copy(&b, r); err != nil {
+		return "", err
+	}
+
+	if contentType := llm.DetectGGMLType(b.Bytes()); contentType != "" {
+		return contentType, nil
+	}
+
+	if contentType := http.DetectContentType(b.Bytes()); contentType != "application/octet-stream" {
+		return contentType, nil
+	}
+
+	return "unknown", nil
+}

server/routes.go

@@ -580,7 +580,7 @@ func (s *Server) CreateModelHandler(c *gin.Context) {
 		ctx, cancel := context.WithCancel(c.Request.Context())
 		defer cancel()
 
-		if err := CreateModel(ctx, model, filepath.Dir(req.Path), req.Quantization, commands, fn); err != nil {
+		if err := CreateModel(ctx, model, filepath.Dir(req.Path), strings.ToUpper(req.Quantization), commands, fn); err != nil {
 			ch <- gin.H{"error": err.Error()}
 		}
 	}()
@@ -728,12 +728,12 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
 		}
 	}
 
-	mf, err := ShowModelfile(model)
-	if err != nil {
-		return nil, err
-	}
-
-	resp.Modelfile = mf
+	var sb strings.Builder
+	fmt.Fprintln(&sb, "# Modelfile generate by \"ollama show\"")
+	fmt.Fprintln(&sb, "# To build a new Modelfile based on this, replace FROM with:")
+	fmt.Fprintf(&sb, "# FROM %s\n\n", model.ShortName)
+	fmt.Fprint(&sb, parser.Format(model.Commands()))
+	resp.Modelfile = sb.String()
 
 	return resp, nil
 }
@@ -810,16 +810,13 @@ func (s *Server) CopyModelHandler(c *gin.Context) {
 
 	src := model.ParseName(r.Source)
 	if !src.IsValid() {
-		_ = c.Error(fmt.Errorf("source %q is invalid", r.Source))
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("source %q is invalid", r.Source)})
+		return
 	}
 
 	dst := model.ParseName(r.Destination)
 	if !dst.IsValid() {
-		_ = c.Error(fmt.Errorf("destination %q is invalid", r.Destination))
-	}
-
-	if len(c.Errors) > 0 {
-		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": c.Errors.Errors()})
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("destination %q is invalid", r.Source)})
 		return
 	}
 
@@ -875,11 +872,6 @@ func (s *Server) CreateBlobHandler(c *gin.Context) {
 		return
 	}
 
-	if _, err := layer.Commit(); err != nil {
-		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
-		return
-	}
-
 	c.Status(http.StatusCreated)
 }
 

server/routes_test.go

@@ -124,14 +124,12 @@ func Test_Routes(t *testing.T) {
 			Method: http.MethodPost,
 			Path:   "/api/create",
 			Setup: func(t *testing.T, req *http.Request) {
-				f, err := os.CreateTemp(t.TempDir(), "ollama-model")
-				assert.Nil(t, err)
-				defer f.Close()
+				fname := createTestFile(t, "ollama-model")
 
 				stream := false
 				createReq := api.CreateRequest{
 					Name:      "t-bone",
-					Modelfile: fmt.Sprintf("FROM %s", f.Name()),
+					Modelfile: fmt.Sprintf("FROM %s", fname),
 					Stream:    &stream,
 				}
 				jsonData, err := json.Marshal(createReq)
@@ -216,28 +214,25 @@ func Test_Routes(t *testing.T) {
 	httpSrv := httptest.NewServer(router)
 	t.Cleanup(httpSrv.Close)
 
-	workDir, err := os.MkdirTemp("", "ollama-test")
-	assert.Nil(t, err)
-	defer os.RemoveAll(workDir)
-	os.Setenv("OLLAMA_MODELS", workDir)
+	t.Setenv("OLLAMA_MODELS", t.TempDir())
 
 	for _, tc := range testCases {
-		t.Logf("Running Test: [%s]", tc.Name)
-		u := httpSrv.URL + tc.Path
-		req, err := http.NewRequestWithContext(context.TODO(), tc.Method, u, nil)
-		assert.Nil(t, err)
+		t.Run(tc.Name, func(t *testing.T) {
+			u := httpSrv.URL + tc.Path
+			req, err := http.NewRequestWithContext(context.TODO(), tc.Method, u, nil)
+			assert.Nil(t, err)
 
-		if tc.Setup != nil {
-			tc.Setup(t, req)
-		}
+			if tc.Setup != nil {
+				tc.Setup(t, req)
+			}
 
-		resp, err := httpSrv.Client().Do(req)
-		assert.Nil(t, err)
-		defer resp.Body.Close()
-
-		if tc.Expected != nil {
-			tc.Expected(t, resp)
-		}
+			resp, err := httpSrv.Client().Do(req)
+			assert.Nil(t, err)
+			defer resp.Body.Close()
 
+			if tc.Expected != nil {
+				tc.Expected(t, resp)
+			}
+		})
 	}
 }