OpenAI Delete Endpoint

Actually skip PhysX on windows

.github/workflows/test.yaml

@@ -34,13 +34,13 @@ jobs:
             git diff-tree -r --no-commit-id --name-only \
               $(git merge-base ${{ github.event.pull_request.base.sha }} ${{ github.event.pull_request.head.sha }}) \
               ${{ github.event.pull_request.head.sha }} \
-              | xargs python3 -c "import sys; print(any([x.startswith('$1') for x in sys.argv[1:]]))"
+              | xargs python3 -c "import sys; from pathlib import Path; print(any(Path(x).match(glob) for x in sys.argv[1:] for glob in '$*'.split(' ')))"
           }
 
           {
-            echo GENERATE=$(changed llm/)
-            echo GENERATE_CUDA=$(changed llm/)
-            echo GENERATE_ROCM=$(changed llm/)
+            echo GENERATE=$(changed 'llm/llama.cpp' 'llm/patches/**' 'llm/ext_server/**' 'llm/generate/**')
+            echo GENERATE_CUDA=$(changed 'llm/llama.cpp' 'llm/patches/**' 'llm/ext_server/**' 'llm/generate/**')
+            echo GENERATE_ROCM=$(changed 'llm/llama.cpp' 'llm/patches/**' 'llm/ext_server/**' 'llm/generate/**')
           } >>$GITHUB_OUTPUT
 
   generate:
@@ -269,9 +269,9 @@ jobs:
           mkdir -p llm/build/darwin/$ARCH/stub/bin
           touch llm/build/darwin/$ARCH/stub/bin/ollama_llama_server
         if: ${{ startsWith(matrix.os, 'macos-') }}
-      - uses: golangci/golangci-lint-action@v4
+      - uses: golangci/golangci-lint-action@v6
         with:
-          args: --timeout 8m0s -v
+          args: --timeout 8m0s -v ${{ startsWith(matrix.os, 'windows-') && '' || '--disable gofmt --disable goimports' }}
   test:
     strategy:
       matrix:
@@ -287,6 +287,8 @@ jobs:
       GOARCH: ${{ matrix.arch }}
       CGO_ENABLED: '1'
       OLLAMA_CPU_TARGET: 'static'
+      OLLAMA_SKIP_CPU_GENERATE: '1'
+      OLLAMA_SKIP_METAL_GENERATE: '1'
     steps:
       - uses: actions/checkout@v4
         with:

.golangci.yaml

@@ -9,9 +9,26 @@ linters:
     - contextcheck
     - exportloopref
     - gocheckcompilerdirectives
-    # FIXME: for some reason this errors on windows
+    # conditionally enable this on linux/macos
     # - gofmt
     # - goimports
+    - intrange
     - misspell
     - nilerr
+    - nolintlint
+    - nosprintfhostport
+    - testifylint
+    - unconvert
     - unused
+    - wastedassign
+    - whitespace
+    - usestdlibvars
+severity:
+  default-severity: error
+  rules:
+    - linters:
+        - gofmt
+        - goimports
+        - intrange
+        - usestdlibvars
+      severity: info

README.md

@@ -6,7 +6,7 @@
 
 [![Discord](https://dcbadge.vercel.app/api/server/ollama?style=flat&compact=true)](https://discord.gg/ollama)
 
-Get up and running with large language models locally.
+Get up and running with large language models.
 
 ### macOS
 
@@ -285,6 +285,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [macai](https://github.com/Renset/macai) (macOS client for Ollama, ChatGPT, and other compatible API back-ends)
 - [Olpaka](https://github.com/Otacon/olpaka) (User-friendly Flutter Web App for Ollama)
 - [OllamaSpring](https://github.com/CrazyNeil/OllamaSpring) (Ollama Client for macOS)
+- [LLocal.in](https://github.com/kartikm7/llocal) (Easy to use Electron Desktop Client for Ollama)
 
 ### Terminal
 
@@ -307,6 +308,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [ShellOracle](https://github.com/djcopley/ShellOracle)
 - [tlm](https://github.com/yusufcanb/tlm)
 - [podman-ollama](https://github.com/ericcurtin/podman-ollama)
+- [gollama](https://github.com/sammcj/gollama)
 
 ### Database
 
@@ -324,11 +326,13 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LangChain](https://python.langchain.com/docs/integrations/llms/ollama) and [LangChain.js](https://js.langchain.com/docs/modules/model_io/models/llms/integrations/ollama) with [example](https://js.langchain.com/docs/use_cases/question_answering/local_retrieval_qa)
 - [LangChainGo](https://github.com/tmc/langchaingo/) with [example](https://github.com/tmc/langchaingo/tree/main/examples/ollama-completion-example)
 - [LangChain4j](https://github.com/langchain4j/langchain4j) with [example](https://github.com/langchain4j/langchain4j-examples/tree/main/ollama-examples/src/main/java)
+- [LangChainRust](https://github.com/Abraxas-365/langchain-rust) with [example](https://github.com/Abraxas-365/langchain-rust/blob/main/examples/llm_ollama.rs)
 - [LlamaIndex](https://gpt-index.readthedocs.io/en/stable/examples/llm/ollama.html)
 - [LiteLLM](https://github.com/BerriAI/litellm)
 - [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
 - [Ollama for Ruby](https://github.com/gbaptista/ollama-ai)
 - [Ollama-rs for Rust](https://github.com/pepperoni21/ollama-rs)
+- [Ollama-hpp for C++](https://github.com/jmont-dev/ollama-hpp)
 - [Ollama4j for Java](https://github.com/amithkoujalgi/ollama4j)
 - [ModelFusion Typescript Library](https://modelfusion.dev/integration/model-provider/ollama)
 - [OllamaKit for Swift](https://github.com/kevinhermawan/OllamaKit)
@@ -346,6 +350,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Portkey](https://portkey.ai/docs/welcome/integration-guides/ollama)
 - [PromptingTools.jl](https://github.com/svilupp/PromptingTools.jl) with an [example](https://svilupp.github.io/PromptingTools.jl/dev/examples/working_with_ollama)
 - [LlamaScript](https://github.com/Project-Llama/llamascript)
+
 ### Mobile
 
 - [Enchanted](https://github.com/AugustDev/enchanted)
@@ -378,7 +383,9 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [AI ST Completion](https://github.com/yaroslavyaroslav/OpenAI-sublime-text) (Sublime Text 4 AI assistant plugin with Ollama support)
 - [Discord-Ollama Chat Bot](https://github.com/kevinthedang/discord-ollama) (Generalized TypeScript Discord Bot w/ Tuning Documentation)
 - [Discord AI chat/moderation bot](https://github.com/rapmd73/Companion) Chat/moderation bot written in python. Uses Ollama to create personalities.
+- [Headless Ollama](https://github.com/nischalj10/headless-ollama) (Scripts to automatically install ollama client & models on any OS for apps that depends on ollama server)
+
+### Supported backends
 
-### Supported backends 
 - [llama.cpp](https://github.com/ggerganov/llama.cpp) project founded by Georgi Gerganov.
 

api/client.go

@@ -23,11 +23,9 @@ import (
 	"net"
 	"net/http"
 	"net/url"
-	"os"
 	"runtime"
-	"strconv"
-	"strings"
 
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/version"
 )
@@ -65,10 +63,7 @@ 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
-	}
+	ollamaHost := envconfig.Host
 
 	return &Client{
 		base: &url.URL{
@@ -79,52 +74,6 @@ func ClientFromEnvironment() (*Client, error) {
 	}, nil
 }
 
-type OllamaHost struct {
-	Scheme string
-	Host   string
-	Port   string
-}
-
-func GetOllamaHost() (OllamaHost, error) {
-	defaultPort := "11434"
-
-	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", hostVar
-	case scheme == "http":
-		defaultPort = "80"
-	case scheme == "https":
-		defaultPort = "443"
-	}
-
-	// trim trailing slashes
-	hostport = strings.TrimRight(hostport, "/")
-
-	host, port, err := net.SplitHostPort(hostport)
-	if err != nil {
-		host, port = "127.0.0.1", defaultPort
-		if ip := net.ParseIP(strings.Trim(hostport, "[]")); ip != nil {
-			host = ip.String()
-		} else if hostport != "" {
-			host = hostport
-		}
-	}
-
-	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
-}
-
 func NewClient(base *url.URL, http *http.Client) *Client {
 	return &Client{
 		base: base,
@@ -355,8 +304,8 @@ func (c *Client) List(ctx context.Context) (*ListResponse, error) {
 }
 
 // List running models.
-func (c *Client) ListRunning(ctx context.Context) (*ListResponse, error) {
-	var lr ListResponse
+func (c *Client) ListRunning(ctx context.Context) (*ProcessResponse, error) {
+	var lr ProcessResponse
 	if err := c.do(ctx, http.MethodGet, "/api/ps", nil, &lr); err != nil {
 		return nil, err
 	}

api/client_test.go

@@ -1,11 +1,9 @@
 package api
 
 import (
-	"fmt"
-	"net"
 	"testing"
 
-	"github.com/stretchr/testify/assert"
+	"github.com/ollama/ollama/envconfig"
 )
 
 func TestClientFromEnvironment(t *testing.T) {
@@ -35,6 +33,7 @@ func TestClientFromEnvironment(t *testing.T) {
 	for k, v := range testCases {
 		t.Run(k, func(t *testing.T) {
 			t.Setenv("OLLAMA_HOST", v.value)
+			envconfig.LoadConfig()
 
 			client, err := ClientFromEnvironment()
 			if err != v.err {
@@ -46,40 +45,4 @@ 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

@@ -2,7 +2,6 @@ package api
 
 import (
 	"encoding/json"
-	"errors"
 	"fmt"
 	"log/slog"
 	"math"
@@ -282,19 +281,33 @@ type PushRequest struct {
 
 // ListResponse is the response from [Client.List].
 type ListResponse struct {
-	Models []ModelResponse `json:"models"`
+	Models []ListModelResponse `json:"models"`
 }
 
-// ModelResponse is a single model description in [ListResponse].
-type ModelResponse struct {
+// ProcessResponse is the response from [Client.Process].
+type ProcessResponse struct {
+	Models []ProcessModelResponse `json:"models"`
+}
+
+// ListModelResponse is a single model description in [ListResponse].
+type ListModelResponse struct {
 	Name       string       `json:"name"`
 	Model      string       `json:"model"`
-	ModifiedAt time.Time    `json:"modified_at,omitempty"`
+	ModifiedAt time.Time    `json:"modified_at"`
 	Size       int64        `json:"size"`
 	Digest     string       `json:"digest"`
 	Details    ModelDetails `json:"details,omitempty"`
-	ExpiresAt  time.Time    `json:"expires_at,omitempty"`
-	SizeVRAM   int64        `json:"size_vram,omitempty"`
+}
+
+// ProcessModelResponse is a single model description in [ProcessResponse].
+type ProcessModelResponse struct {
+	Name      string       `json:"name"`
+	Model     string       `json:"model"`
+	Size      int64        `json:"size"`
+	Digest    string       `json:"digest"`
+	Details   ModelDetails `json:"details,omitempty"`
+	ExpiresAt time.Time    `json:"expires_at"`
+	SizeVRAM  int64        `json:"size_vram"`
 }
 
 type TokenResponse struct {
@@ -306,7 +319,7 @@ type GenerateResponse struct {
 	// Model is the model name that generated the response.
 	Model string `json:"model"`
 
-	//CreatedAt is the timestamp of the response.
+	// CreatedAt is the timestamp of the response.
 	CreatedAt time.Time `json:"created_at"`
 
 	// Response is the textual response itself.
@@ -363,8 +376,6 @@ func (m *Metrics) Summary() {
 	}
 }
 
-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
 	typeOpts := reflect.TypeOf(opts).Elem()   // types of the fields in the options struct

api/types_test.go

@@ -72,13 +72,13 @@ func TestDurationMarshalUnmarshal(t *testing.T) {
 		},
 		{
 			"positive duration",
-			time.Duration(42 * time.Second),
-			time.Duration(42 * time.Second),
+			42 * time.Second,
+			42 * time.Second,
 		},
 		{
 			"another positive duration",
-			time.Duration(42 * time.Minute),
-			time.Duration(42 * time.Minute),
+			42 * time.Minute,
+			42 * time.Minute,
 		},
 		{
 			"zero duration",

app/lifecycle/paths.go

@@ -69,7 +69,6 @@ func init() {
 				slog.Error(fmt.Sprintf("create ollama dir %s: %v", AppDataDir, err))
 			}
 		}
-
 	} else if runtime.GOOS == "darwin" {
 		// TODO
 		AppName += ".app"

app/lifecycle/server.go

@@ -15,7 +15,7 @@ import (
 )
 
 func getCLIFullPath(command string) string {
-	cmdPath := ""
+	var cmdPath string
 	appExe, err := os.Executable()
 	if err == nil {
 		cmdPath = filepath.Join(filepath.Dir(appExe), command)
@@ -65,7 +65,6 @@ func start(ctx context.Context, command string) (*exec.Cmd, error) {
 	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 {

app/lifecycle/server_windows.go

@@ -24,7 +24,8 @@ func terminate(cmd *exec.Cmd) error {
 	if err != nil {
 		return err
 	}
-	defer dll.Release() // nolint: errcheck
+	//nolint:errcheck
+	defer dll.Release()
 
 	pid := cmd.Process.Pid
 
@@ -73,7 +74,8 @@ func isProcessExited(pid int) (bool, error) {
 	if err != nil {
 		return false, fmt.Errorf("failed to open process: %v", err)
 	}
-	defer windows.CloseHandle(hProcess) // nolint: errcheck
+	//nolint:errcheck
+	defer windows.CloseHandle(hProcess)
 
 	var exitCode uint32
 	err = windows.GetExitCodeProcess(hProcess, &exitCode)

app/lifecycle/updater.go

@@ -78,7 +78,7 @@ func IsNewReleaseAvailable(ctx context.Context) (bool, UpdateResponse) {
 	}
 	defer resp.Body.Close()
 
-	if resp.StatusCode == 204 {
+	if resp.StatusCode == http.StatusNoContent {
 		slog.Debug("check update response 204 (current version is up to date)")
 		return false, updateResp
 	}
@@ -87,7 +87,7 @@ func IsNewReleaseAvailable(ctx context.Context) (bool, UpdateResponse) {
 		slog.Warn(fmt.Sprintf("failed to read body response: %s", err))
 	}
 
-	if resp.StatusCode != 200 {
+	if resp.StatusCode != http.StatusOK {
 		slog.Info(fmt.Sprintf("check update error %d - %.96s", resp.StatusCode, string(body)))
 		return false, updateResp
 	}
@@ -114,7 +114,7 @@ func DownloadNewRelease(ctx context.Context, updateResp UpdateResponse) error {
 	if err != nil {
 		return fmt.Errorf("error checking update: %w", err)
 	}
-	if resp.StatusCode != 200 {
+	if resp.StatusCode != http.StatusOK {
 		return fmt.Errorf("unexpected status attempting to download update %d", resp.StatusCode)
 	}
 	resp.Body.Close()

app/ollama_welcome.ps1

@@ -4,5 +4,5 @@ write-host "Welcome to Ollama!"
 write-host ""
 write-host "Run your first model:"
 write-host ""
-write-host "`tollama run llama2"
+write-host "`tollama run llama3"
 write-host ""

app/store/store.go

@@ -29,7 +29,6 @@ func GetID() string {
 		initStore()
 	}
 	return store.ID
-
 }
 
 func GetFirstTimeRun() bool {

app/tray/wintray/eventloop.go

@@ -47,7 +47,6 @@ func nativeLoop() {
 		default:
 			pTranslateMessage.Call(uintptr(unsafe.Pointer(m))) //nolint:errcheck
 			pDispatchMessage.Call(uintptr(unsafe.Pointer(m)))  //nolint:errcheck
-
 		}
 	}
 }
@@ -160,8 +159,8 @@ func (t *winTray) wndProc(hWnd windows.Handle, message uint32, wParam, lParam ui
 		lResult, _, _ = pDefWindowProc.Call(
 			uintptr(hWnd),
 			uintptr(message),
-			uintptr(wParam),
-			uintptr(lParam),
+			wParam,
+			lParam,
 		)
 	}
 	return

app/tray/wintray/tray.go

@@ -186,7 +186,7 @@ func (t *winTray) initInstance() error {
 	t.muNID.Lock()
 	defer t.muNID.Unlock()
 	t.nid = &notifyIconData{
-		Wnd:             windows.Handle(t.window),
+		Wnd:             t.window,
 		ID:              100,
 		Flags:           NIF_MESSAGE,
 		CallbackMessage: t.wmSystrayMessage,
@@ -197,7 +197,6 @@ func (t *winTray) initInstance() error {
 }
 
 func (t *winTray) createMenu() error {
-
 	menuHandle, _, err := pCreatePopupMenu.Call()
 	if menuHandle == 0 {
 		return err
@@ -246,7 +245,7 @@ func (t *winTray) addOrUpdateMenuItem(menuItemId uint32, parentId uint32, title
 	mi := menuItemInfo{
 		Mask:     MIIM_FTYPE | MIIM_STRING | MIIM_ID | MIIM_STATE,
 		Type:     MFT_STRING,
-		ID:       uint32(menuItemId),
+		ID:       menuItemId,
 		TypeData: titlePtr,
 		Cch:      uint32(len(title)),
 	}
@@ -302,11 +301,10 @@ func (t *winTray) addOrUpdateMenuItem(menuItemId uint32, parentId uint32, title
 }
 
 func (t *winTray) addSeparatorMenuItem(menuItemId, parentId uint32) error {
-
 	mi := menuItemInfo{
 		Mask: MIIM_FTYPE | MIIM_ID | MIIM_STATE,
 		Type: MFT_SEPARATOR,
-		ID:   uint32(menuItemId),
+		ID:   menuItemId,
 	}
 
 	mi.Size = uint32(unsafe.Sizeof(mi))
@@ -426,7 +424,6 @@ func iconBytesToFilePath(iconBytes []byte) (string, error) {
 // Loads an image from file and shows it in tray.
 // Shell_NotifyIcon: https://msdn.microsoft.com/en-us/library/windows/desktop/bb762159(v=vs.85).aspx
 func (t *winTray) setIcon(src string) error {
-
 	h, err := t.loadIconFrom(src)
 	if err != nil {
 		return err
@@ -444,7 +441,6 @@ func (t *winTray) setIcon(src string) error {
 // Loads an image from file to be shown in tray or menu item.
 // LoadImage: https://msdn.microsoft.com/en-us/library/windows/desktop/ms648045(v=vs.85).aspx
 func (t *winTray) loadIconFrom(src string) (windows.Handle, error) {
-
 	// Save and reuse handles of loaded images
 	t.muLoadedImages.RLock()
 	h, ok := t.loadedImages[src]

cmd/cmd.go

@@ -20,6 +20,7 @@ import (
 	"path/filepath"
 	"regexp"
 	"runtime"
+	"slices"
 	"strings"
 	"syscall"
 	"time"
@@ -29,7 +30,6 @@ import (
 	"github.com/olekukonko/tablewriter"
 	"github.com/spf13/cobra"
 	"golang.org/x/crypto/ssh"
-	"golang.org/x/exp/slices"
 	"golang.org/x/term"
 
 	"github.com/ollama/ollama/api"
@@ -746,7 +746,6 @@ func displayResponse(content string, wordWrap bool, state *displayResponseState)
 	if wordWrap && termWidth >= 10 {
 		for _, ch := range content {
 			if state.lineLength+1 > termWidth-5 {
-
 				if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
 					fmt.Printf("%s%c", state.wordBuffer, ch)
 					state.wordBuffer = ""
@@ -755,7 +754,11 @@ func displayResponse(content string, wordWrap bool, state *displayResponseState)
 				}
 
 				// backtrack the length of the last word and clear to the end of the line
-				fmt.Printf("\x1b[%dD\x1b[K\n", runewidth.StringWidth(state.wordBuffer))
+				a := runewidth.StringWidth(state.wordBuffer)
+				if a > 0 {
+					fmt.Printf("\x1b[%dD", a)
+				}
+				fmt.Printf("\x1b[K\n")
 				fmt.Printf("%s%c", state.wordBuffer, ch)
 				chWidth := runewidth.RuneWidth(ch)
 
@@ -957,17 +960,11 @@ func generate(cmd *cobra.Command, opts runOptions) error {
 }
 
 func RunServer(cmd *cobra.Command, _ []string) error {
-	// retrieve the OLLAMA_HOST environment variable
-	ollamaHost, err := api.GetOllamaHost()
-	if err != nil {
-		return err
-	}
-
 	if err := initializeKeypair(); err != nil {
 		return err
 	}
 
-	ln, err := net.Listen("tcp", net.JoinHostPort(ollamaHost.Host, ollamaHost.Port))
+	ln, err := net.Listen("tcp", net.JoinHostPort(envconfig.Host.Host, envconfig.Host.Port))
 	if err != nil {
 		return err
 	}
@@ -1026,24 +1023,6 @@ func initializeKeypair() error {
 	return nil
 }
 
-//nolint:unused
-func waitForServer(ctx context.Context, client *api.Client) error {
-	// wait for the server to start
-	timeout := time.After(5 * time.Second)
-	tick := time.Tick(500 * time.Millisecond)
-	for {
-		select {
-		case <-timeout:
-			return errors.New("timed out waiting for server to start")
-		case <-tick:
-			if err := client.Heartbeat(ctx); err == nil {
-				return nil // server has started
-			}
-		}
-	}
-
-}
-
 func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
@@ -1251,6 +1230,9 @@ func NewCLI() *cobra.Command {
 				envVars["OLLAMA_NOPRUNE"],
 				envVars["OLLAMA_ORIGINS"],
 				envVars["OLLAMA_TMPDIR"],
+				envVars["OLLAMA_FLASH_ATTENTION"],
+				envVars["OLLAMA_LLM_LIBRARY"],
+				envVars["OLLAMA_MAX_VRAM"],
 			})
 		default:
 			appendEnvDocs(cmd, envs)

cmd/interactive.go

@@ -8,11 +8,11 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"slices"
 	"sort"
 	"strings"
 
 	"github.com/spf13/cobra"
-	"golang.org/x/exp/slices"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/envconfig"

cmd/interactive_test.go

@@ -6,6 +6,7 @@ import (
 	"text/template"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 
 	"github.com/ollama/ollama/api"
 )
@@ -85,11 +86,11 @@ MESSAGE assistant """Yes it is true, I am half horse, half shark."""
 `
 
 	tmpl, err := template.New("").Parse(expectedModelfile)
-	assert.Nil(t, err)
+	require.NoError(t, err)
 
 	var buf bytes.Buffer
 	err = tmpl.Execute(&buf, opts)
-	assert.Nil(t, err)
+	require.NoError(t, err)
 	assert.Equal(t, buf.String(), mf)
 
 	opts.ParentModel = "horseshark"
@@ -107,10 +108,10 @@ MESSAGE assistant """Yes it is true, I am half horse, half shark."""
 `
 
 	tmpl, err = template.New("").Parse(expectedModelfile)
-	assert.Nil(t, err)
+	require.NoError(t, err)
 
 	var parentBuf bytes.Buffer
 	err = tmpl.Execute(&parentBuf, opts)
-	assert.Nil(t, err)
+	require.NoError(t, err)
 	assert.Equal(t, parentBuf.String(), mf)
 }

cmd/start.go

@@ -0,0 +1,27 @@
+//go:build darwin || windows
+
+package cmd
+
+import (
+	"context"
+	"errors"
+	"time"
+
+	"github.com/ollama/ollama/api"
+)
+
+func waitForServer(ctx context.Context, client *api.Client) error {
+	// wait for the server to start
+	timeout := time.After(5 * time.Second)
+	tick := time.Tick(500 * time.Millisecond)
+	for {
+		select {
+		case <-timeout:
+			return errors.New("timed out waiting for server to start")
+		case <-tick:
+			if err := client.Heartbeat(ctx); err == nil {
+				return nil // server has started
+			}
+		}
+	}
+}

convert/convert.go

@@ -189,7 +189,7 @@ func LoadSentencePieceTokens(dirpath string, params *Params) (*Vocab, error) {
 	if params.VocabSize > len(v.Tokens) {
 		missingTokens := params.VocabSize - len(v.Tokens)
 		slog.Warn(fmt.Sprintf("vocab is missing %d tokens", missingTokens))
-		for cnt := 0; cnt < missingTokens; cnt++ {
+		for cnt := range missingTokens {
 			v.Tokens = append(v.Tokens, fmt.Sprintf("<dummy%05d>", cnt+1))
 			v.Scores = append(v.Scores, -1)
 			v.Types = append(v.Types, tokenTypeUserDefined)

convert/gemma.go

@@ -35,7 +35,6 @@ func addOnes(data []float32, vectorSize int) ([]float32, error) {
 		f32s = append(f32s, t...)
 	}
 
-
 	return f32s, nil
 }
 

convert/llama.go

@@ -119,11 +119,12 @@ func llamaRepack(name string, params *Params, data []float32, shape []uint64) ([
 	}
 
 	var heads int
-	if strings.HasSuffix(name, "attn_q.weight") {
+	switch {
+	case strings.HasSuffix(name, "attn_q.weight"):
 		heads = params.AttentionHeads
-	} else if strings.HasSuffix(name, "attn_k.weight") {
+	case strings.HasSuffix(name, "attn_k.weight"):
 		heads = cmp.Or(params.KeyValHeads, params.AttentionHeads)
-	} else {
+	default:
 		return nil, fmt.Errorf("unknown tensor name: %s", name)
 	}
 

convert/safetensors.go

@@ -120,7 +120,7 @@ func (m *SafetensorFormat) readTensors(fn string, offset uint64, params *Params)
 			Name:   name,
 			Kind:   kind,
 			Offset: offset,
-			Shape:  shape[:],
+			Shape:  shape,
 		}
 
 		t.WriterTo = safetensorWriterTo{

convert/tokenizer.go

@@ -85,11 +85,8 @@ func parseTokens(dirpath string) (pre string, tokens []Token, merges []string, e
 
 	sha256sum := sha256.New()
 	for _, pt := range t.PreTokenizer.PreTokenizers {
-		switch pt.Type {
-		case "Split":
-			if pt.Pattern.Regex != "" {
-				sha256sum.Write([]byte(pt.Pattern.Regex))
-			}
+		if pt.Type == "Split" && pt.Pattern.Regex != "" {
+			sha256sum.Write([]byte(pt.Pattern.Regex))
 		}
 	}
 

convert/torch.go

@@ -88,7 +88,7 @@ func (tf *TorchFormat) GetTensors(dirpath string, params *Params) ([]llm.Tensor,
 				Name:   ggufName,
 				Kind:   kind,
 				Offset: offset, // calculate the offset
-				Shape:  shape[:],
+				Shape:  shape,
 			}
 
 			tensor.WriterTo = torchWriterTo{
@@ -104,7 +104,6 @@ func (tf *TorchFormat) GetTensors(dirpath string, params *Params) ([]llm.Tensor,
 	}
 
 	return tensors, nil
-
 }
 
 func getAltParams(dirpath string) (*Params, error) {

docs/api.md

@@ -12,6 +12,7 @@
 - [Pull a Model](#pull-a-model)
 - [Push a Model](#push-a-model)
 - [Generate Embeddings](#generate-embeddings)
+- [List Running Models](#list-running-models)
 
 ## Conventions
 
@@ -249,7 +250,7 @@ curl http://localhost:11434/api/generate -d '{
 
 #### Request (Reproducible outputs)
 
-For reproducible outputs, set `temperature` to 0 and `seed` to a number:
+For reproducible outputs, set `seed` to a number:
 
 ##### Request
 
@@ -258,8 +259,7 @@ curl http://localhost:11434/api/generate -d '{
   "model": "mistral",
   "prompt": "Why is the sky blue?",
   "options": {
-    "seed": 123,
-    "temperature": 0
+    "seed": 123
   }
 }'
 ```
@@ -1035,3 +1035,47 @@ curl http://localhost:11434/api/embeddings -d '{
   ]
 }
 ```
+
+## List Running Models
+```shell
+GET /api/ps
+```
+
+List models that are currently loaded into memory.
+
+#### Examples
+
+### Request
+
+```shell
+curl http://localhost:11434/api/ps
+```
+
+#### Response
+
+A single JSON object will be returned.
+
+```json
+{
+  "models": [
+    {
+      "name": "mistral:latest",
+      "model": "mistral:latest",
+      "size": 5137025024,
+      "digest": "2ae6f6dd7a3dd734790bbbf58b8909a606e0e7e97e94b7604e0aa7ae4490e6d8",
+      "details": {
+        "parent_model": "",
+        "format": "gguf",
+        "family": "llama",
+        "families": [
+          "llama"
+        ],
+        "parameter_size": "7.2B",
+        "quantization_level": "Q4_0"
+      },
+      "expires_at": "2024-06-04T14:38:31.83753-07:00",
+      "size_vram": 5137025024
+    }
+  ]
+}
+```

docs/import.md

@@ -1,170 +1,99 @@
-# Import a model
+# Import
 
-This guide walks through importing a GGUF, PyTorch or Safetensors model.
+GGUF models and select Safetensors models can be imported directly into Ollama.
 
-## Importing (GGUF)
+## Import GGUF
 
-### Step 1: Write a `Modelfile`
+A binary GGUF file can be imported directly into Ollama through a Modelfile.
 
-Start by creating a `Modelfile`. This file is the blueprint for your model, specifying weights, parameters, prompt templates and more.
-
-```
-FROM ./mistral-7b-v0.1.Q4_0.gguf
+```dockerfile
+FROM /path/to/file.gguf
 ```
 
-(Optional) many chat models require a prompt template in order to answer correctly. A default prompt template can be specified with the `TEMPLATE` instruction in the `Modelfile`:
+## Import Safetensors
 
-```
-FROM ./mistral-7b-v0.1.Q4_0.gguf
-TEMPLATE "[INST] {{ .Prompt }} [/INST]"
+If the model being imported is one of these architectures, it can be imported directly into Ollama through a Modelfile:
+
+ - LlamaForCausalLM
+ - MistralForCausalLM
+ - GemmaForCausalLM
+
+```dockerfile
+FROM /path/to/safetensors/directory
 ```
 
-### Step 2: Create the Ollama model
+For architectures not directly convertable by Ollama, see llama.cpp's [guide](https://github.com/ggerganov/llama.cpp/blob/master/README.md#prepare-and-quantize) on conversion. After conversion, see [Import GGUF](#import-gguf).
 
-Finally, create a model from your `Modelfile`:
+## Automatic Quantization
 
+> [!NOTE]
+> Automatic quantization requires v0.1.35 or higher.
+
+Ollama is capable of quantizing FP16 or FP32 models to any of the supported quantizations with the `-q/--quantize` flag in `ollama create`.
+
+```dockerfile
+FROM /path/to/my/gemma/f16/model
 ```
-ollama create example -f Modelfile
-```
-
-### Step 3: Run your model
-
-Next, test the model with `ollama run`:
-
-```
-ollama run example "What is your favourite condiment?"
-```
-
-## Importing (PyTorch & Safetensors)
-
-> Importing from PyTorch and Safetensors is a longer process than importing from GGUF. Improvements that make it easier are a work in progress.
-
-### Setup
-
-First, clone the `ollama/ollama` repo:
-
-```
-git clone git@github.com:ollama/ollama.git ollama
-cd ollama
-```
-
-and then fetch its `llama.cpp` submodule:
 
 ```shell
-git submodule init
-git submodule update llm/llama.cpp
+$ ollama create -q Q4_K_M mymodel
+transferring model data
+quantizing F16 model to Q4_K_M
+creating new layer sha256:735e246cc1abfd06e9cdcf95504d6789a6cd1ad7577108a70d9902fef503c1bd
+creating new layer sha256:0853f0ad24e5865173bbf9ffcc7b0f5d56b66fd690ab1009867e45e7d2c4db0f
+writing manifest
+success
 ```
 
-Next, install the Python dependencies:
+### Supported Quantizations
 
-```
-python3 -m venv llm/llama.cpp/.venv
-source llm/llama.cpp/.venv/bin/activate
-pip install -r llm/llama.cpp/requirements.txt
+<details>
+<summary>Legacy Quantization</summary>
+
+- `Q4_0`
+- `Q4_1`
+- `Q5_0`
+- `Q5_1`
+- `Q8_0`
+
+</details>
+
+<details>
+<summary>K-means Quantization</summary>`
+
+- `Q3_K_S`
+- `Q3_K_M`
+- `Q3_K_L`
+- `Q4_K_S`
+- `Q4_K_M`
+- `Q5_K_S`
+- `Q5_K_M`
+- `Q6_K`
+
+</details>
+
+> [!NOTE]
+> Activation-aware Weight Quantization (i.e. IQ) are not currently supported for automatic quantization however you can still import the quantized model into Ollama, see [Import GGUF](#import-gguf).
+
+## Template Detection
+
+> [!NOTE]
+> Template detection requires v0.1.42 or higher.
+
+Ollama uses model metadata, specifically `tokenizer.chat_template`, to automatically create a template appropriate for the model you're importing.
+
+```dockerfile
+FROM /path/to/my/gemma/model
 ```
 
-Then build the `quantize` tool:
-
-```
-make -C llm/llama.cpp quantize
+```shell
+$ ollama create mymodel
+transferring model data
+using autodetected template gemma-instruct
+creating new layer sha256:baa2a0edc27d19cc6b7537578a9a7ba1a4e3214dc185ed5ae43692b319af7b84
+creating new layer sha256:ba66c3309914dbef07e5149a648fd1877f030d337a4f240d444ea335008943cb
+writing manifest
+success
 ```
 
-### Clone the HuggingFace repository (optional)
-
-If the model is currently hosted in a HuggingFace repository, first clone that repository to download the raw model.
-
-Install [Git LFS](https://docs.github.com/en/repositories/working-with-files/managing-large-files/installing-git-large-file-storage), verify it's installed, and then clone the model's repository:
-
-```
-git lfs install
-git clone https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1 model
-```
-
-### Convert the model
-
-> Note: some model architectures require using specific convert scripts. For example, Qwen models require running `convert-hf-to-gguf.py` instead of `convert.py`
-
-```
-python llm/llama.cpp/convert.py ./model --outtype f16 --outfile converted.bin
-```
-
-### Quantize the model
-
-```
-llm/llama.cpp/quantize converted.bin quantized.bin q4_0
-```
-
-### Step 3: Write a `Modelfile`
-
-Next, create a `Modelfile` for your model:
-
-```
-FROM quantized.bin
-TEMPLATE "[INST] {{ .Prompt }} [/INST]"
-```
-
-### Step 4: Create the Ollama model
-
-Finally, create a model from your `Modelfile`:
-
-```
-ollama create example -f Modelfile
-```
-
-### Step 5: Run your model
-
-Next, test the model with `ollama run`:
-
-```
-ollama run example "What is your favourite condiment?"
-```
-
-## Publishing your model (optional – early alpha)
-
-Publishing models is in early alpha. If you'd like to publish your model to share with others, follow these steps:
-
-1. Create [an account](https://ollama.com/signup)
-2. Copy your Ollama public key:
-  - macOS: `cat ~/.ollama/id_ed25519.pub | pbcopy`
-  - Windows: `type %USERPROFILE%\.ollama\id_ed25519.pub`
-  - Linux: `cat /usr/share/ollama/.ollama/id_ed25519.pub`
-3. Add your public key to your [Ollama account](https://ollama.com/settings/keys)
-
-Next, copy your model to your username's namespace:
-
-```
-ollama cp example <your username>/example
-```
-
-> Note: model names may only contain lowercase letters, digits, and the characters `.`, `-`, and `_`.
-
-Then push the model:
-
-```
-ollama push <your username>/example
-```
-
-After publishing, your model will be available at `https://ollama.com/<your username>/example`.
-
-## Quantization reference
-
-The quantization options are as follow (from highest highest to lowest levels of quantization). Note: some architectures such as Falcon do not support K quants.
-
-- `q2_K`
-- `q3_K`
-- `q3_K_S`
-- `q3_K_M`
-- `q3_K_L`
-- `q4_0` (recommended)
-- `q4_1`
-- `q4_K`
-- `q4_K_S`
-- `q4_K_M`
-- `q5_0`
-- `q5_1`
-- `q5_K`
-- `q5_K_S`
-- `q5_K_M`
-- `q6_K`
-- `q8_0`
-- `f16`
+Defining a template in the Modelfile will disable this feature which may be useful if you want to use a different template than the autodetected one.

docs/linux.md

@@ -100,6 +100,16 @@ sudo curl -L https://ollama.com/download/ollama-linux-amd64 -o /usr/bin/ollama
 sudo chmod +x /usr/bin/ollama
 ```
 
+## Installing specific versions
+
+Use `OLLAMA_VERSION` environment variable with the install script to install a specific version of Ollama, including pre-releases. You can find the version numbers in the [releases page](https://github.com/ollama/ollama/releases). 
+
+For example:
+
+```
+curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.1.32 sh
+```
+
 ## Viewing logs
 
 To view logs of Ollama running as a startup service, run:

docs/troubleshooting.md

@@ -76,6 +76,7 @@ Make sure you've set up the container runtime first as described in [docker.md](
 
 Sometimes the container runtime can have difficulties initializing the GPU. When you check the server logs, this can show up as various error codes, such as "3" (not initialized), "46" (device unavailable), "100" (no device), "999" (unknown), or others. The following troubleshooting techniques may help resolve the problem
 
+- Is the container runtime working?  Try `docker run --gpus all ubuntu nvidia-smi` - if this doesn't work, Ollama wont be able to see your NVIDIA GPU.
 - Is the uvm driver not loaded? `sudo nvidia-modprobe -u`
 - Try reloading the nvidia_uvm driver - `sudo rmmod nvidia_uvm` then `sudo modprobe nvidia_uvm`
 - Try rebooting

docs/tutorials/langchainpy.md

@@ -45,7 +45,7 @@ all_splits = text_splitter.split_documents(data)
 ```
 
 It's split up, but we have to find the relevant splits and then submit those to the model. We can do this by creating embeddings and storing them in a vector database. We can use Ollama directly to instantiate an embedding model. We will use ChromaDB in this example for a vector database. `pip install chromadb`
-
+We also need to pull embedding model: `ollama pull nomic-embed-text`
 ```python
 from langchain.embeddings import OllamaEmbeddings
 from langchain.vectorstores import Chroma
@@ -68,7 +68,8 @@ The next thing is to send the question and the relevant parts of the docs to the
 ```python
 from langchain.chains import RetrievalQA
 qachain=RetrievalQA.from_chain_type(ollama, retriever=vectorstore.as_retriever())
-qachain.invoke({"query": question})
+res = qachain.invoke({"query": question})
+print(res['result'])
 ```
 
 The answer received from this chain was:

envconfig/config.go

@@ -1,8 +1,10 @@
 package envconfig
 
 import (
+	"errors"
 	"fmt"
 	"log/slog"
+	"net"
 	"os"
 	"path/filepath"
 	"runtime"
@@ -10,6 +12,18 @@ import (
 	"strings"
 )
 
+type OllamaHost struct {
+	Scheme string
+	Host   string
+	Port   string
+}
+
+func (o OllamaHost) String() string {
+	return fmt.Sprintf("%s://%s:%s", o.Scheme, o.Host, o.Port)
+}
+
+var ErrInvalidHostPort = errors.New("invalid port specified in OLLAMA_HOST")
+
 var (
 	// Set via OLLAMA_ORIGINS in the environment
 	AllowOrigins []string
@@ -17,6 +31,8 @@ var (
 	Debug bool
 	// Experimental flash attention
 	FlashAttention bool
+	// Set via OLLAMA_HOST in the environment
+	Host *OllamaHost
 	// Set via OLLAMA_KEEP_ALIVE in the environment
 	KeepAlive string
 	// Set via OLLAMA_LLM_LIBRARY in the environment
@@ -25,6 +41,8 @@ var (
 	MaxRunners int
 	// Set via OLLAMA_MAX_QUEUE in the environment
 	MaxQueuedRequests int
+	// Set via OLLAMA_MODELS in the environment
+	ModelsDir string
 	// Set via OLLAMA_MAX_VRAM in the environment
 	MaxVRAM uint64
 	// Set via OLLAMA_NOHISTORY in the environment
@@ -49,18 +67,18 @@ func AsMap() map[string]EnvVar {
 	return map[string]EnvVar{
 		"OLLAMA_DEBUG":             {"OLLAMA_DEBUG", Debug, "Show additional debug information (e.g. OLLAMA_DEBUG=1)"},
 		"OLLAMA_FLASH_ATTENTION":   {"OLLAMA_FLASH_ATTENTION", FlashAttention, "Enabled flash attention"},
-		"OLLAMA_HOST":              {"OLLAMA_HOST", "", "IP Address for the ollama server (default 127.0.0.1:11434)"},
+		"OLLAMA_HOST":              {"OLLAMA_HOST", Host, "IP Address for the ollama server (default 127.0.0.1:11434)"},
 		"OLLAMA_KEEP_ALIVE":        {"OLLAMA_KEEP_ALIVE", KeepAlive, "The duration that models stay loaded in memory (default \"5m\")"},
-		"OLLAMA_LLM_LIBRARY":       {"OLLAMA_ORIGINS", LLMLibrary, ""},
+		"OLLAMA_LLM_LIBRARY":       {"OLLAMA_LLM_LIBRARY", LLMLibrary, "Set LLM library to bypass autodetection"},
 		"OLLAMA_MAX_LOADED_MODELS": {"OLLAMA_MAX_LOADED_MODELS", MaxRunners, "Maximum number of loaded models (default 1)"},
 		"OLLAMA_MAX_QUEUE":         {"OLLAMA_MAX_QUEUE", MaxQueuedRequests, "Maximum number of queued requests"},
-		"OLLAMA_MAX_VRAM":          {"OLLAMA_MAX_VRAM", MaxVRAM, ""},
-		"OLLAMA_MODELS":            {"OLLAMA_MODELS", "", "The path to the models directory"},
+		"OLLAMA_MAX_VRAM":          {"OLLAMA_MAX_VRAM", MaxVRAM, "Maximum VRAM"},
+		"OLLAMA_MODELS":            {"OLLAMA_MODELS", ModelsDir, "The path to the models directory"},
 		"OLLAMA_NOHISTORY":         {"OLLAMA_NOHISTORY", NoHistory, "Do not preserve readline history"},
 		"OLLAMA_NOPRUNE":           {"OLLAMA_NOPRUNE", NoPrune, "Do not prune model blobs on startup"},
 		"OLLAMA_NUM_PARALLEL":      {"OLLAMA_NUM_PARALLEL", NumParallel, "Maximum number of parallel requests (default 1)"},
 		"OLLAMA_ORIGINS":           {"OLLAMA_ORIGINS", AllowOrigins, "A comma separated list of allowed origins"},
-		"OLLAMA_RUNNERS_DIR":       {"OLLAMA_RUNNERS_DIR", RunnersDir, ""},
+		"OLLAMA_RUNNERS_DIR":       {"OLLAMA_RUNNERS_DIR", RunnersDir, "Location for runners"},
 		"OLLAMA_TMPDIR":            {"OLLAMA_TMPDIR", TmpDir, "Location for temporary files"},
 	}
 }
@@ -126,7 +144,7 @@ func LoadConfig() {
 		var paths []string
 		for _, root := range []string{filepath.Dir(appExe), cwd} {
 			paths = append(paths,
-				filepath.Join(root),
+				root,
 				filepath.Join(root, "windows-"+runtime.GOARCH),
 				filepath.Join(root, "dist", "windows-"+runtime.GOARCH),
 			)
@@ -184,11 +202,17 @@ func LoadConfig() {
 		AllowOrigins = append(AllowOrigins,
 			fmt.Sprintf("http://%s", allowOrigin),
 			fmt.Sprintf("https://%s", allowOrigin),
-			fmt.Sprintf("http://%s:*", allowOrigin),
-			fmt.Sprintf("https://%s:*", allowOrigin),
+			fmt.Sprintf("http://%s", net.JoinHostPort(allowOrigin, "*")),
+			fmt.Sprintf("https://%s", net.JoinHostPort(allowOrigin, "*")),
 		)
 	}
 
+	AllowOrigins = append(AllowOrigins,
+		"app://*",
+		"file://*",
+		"tauri://*",
+	)
+
 	maxRunners := clean("OLLAMA_MAX_LOADED_MODELS")
 	if maxRunners != "" {
 		m, err := strconv.Atoi(maxRunners)
@@ -209,4 +233,70 @@ func LoadConfig() {
 	}
 
 	KeepAlive = clean("OLLAMA_KEEP_ALIVE")
+
+	var err error
+	ModelsDir, err = getModelsDir()
+	if err != nil {
+		slog.Error("invalid setting", "OLLAMA_MODELS", ModelsDir, "error", err)
+	}
+
+	Host, err = getOllamaHost()
+	if err != nil {
+		slog.Error("invalid setting", "OLLAMA_HOST", Host, "error", err, "using default port", Host.Port)
+	}
+}
+
+func getModelsDir() (string, error) {
+	if models, exists := os.LookupEnv("OLLAMA_MODELS"); exists {
+		return models, nil
+	}
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return "", err
+	}
+	return filepath.Join(home, ".ollama", "models"), nil
+}
+
+func getOllamaHost() (*OllamaHost, error) {
+	defaultPort := "11434"
+
+	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", hostVar
+	case scheme == "http":
+		defaultPort = "80"
+	case scheme == "https":
+		defaultPort = "443"
+	}
+
+	// trim trailing slashes
+	hostport = strings.TrimRight(hostport, "/")
+
+	host, port, err := net.SplitHostPort(hostport)
+	if err != nil {
+		host, port = "127.0.0.1", defaultPort
+		if ip := net.ParseIP(strings.Trim(hostport, "[]")); ip != nil {
+			host = ip.String()
+		} else if hostport != "" {
+			host = hostport
+		}
+	}
+
+	if portNum, err := strconv.ParseInt(port, 10, 32); err != nil || portNum > 65535 || portNum < 0 {
+		return &OllamaHost{
+			Scheme: scheme,
+			Host:   host,
+			Port:   defaultPort,
+		}, ErrInvalidHostPort
+	}
+
+	return &OllamaHost{
+		Scheme: scheme,
+		Host:   host,
+		Port:   port,
+	}, nil
 }

envconfig/config_test.go

@@ -1,8 +1,11 @@
 package envconfig
 
 import (
+	"fmt"
+	"net"
 	"testing"
 
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
@@ -21,3 +24,48 @@ func TestConfig(t *testing.T) {
 	LoadConfig()
 	require.True(t, FlashAttention)
 }
+
+func TestClientFromEnvironment(t *testing.T) {
+	type testCase struct {
+		value  string
+		expect string
+		err    error
+	}
+
+	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)
+			LoadConfig()
+
+			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))
+			}
+		})
+	}
+}

examples/langchain-python-rag-privategpt/ingest.py

@@ -77,13 +77,21 @@ LOADER_MAPPING = {
 
 
 def load_single_document(file_path: str) -> List[Document]:
-    ext = "." + file_path.rsplit(".", 1)[-1]
-    if ext in LOADER_MAPPING:
-        loader_class, loader_args = LOADER_MAPPING[ext]
-        loader = loader_class(file_path, **loader_args)
-        return loader.load()
+    if os.path.getsize(file_path) != 0:
+        filename, ext = os.path.splitext(file_path)
+        if ext in LOADER_MAPPING:
+            loader_class, loader_args = LOADER_MAPPING[ext]
+            try:
+                loader = loader_class(file_path, **loader_args)
+                if loader:
+                    return loader.load()
+            except:
+                print(f"Corrupted file {file_path}. Ignoring it.")
+        else:
+            print(f"Unsupported file {file_path}. Ignoring it.")
+    else:
+        print(f"Empty file {file_path}. Ignoring it.")
 
-    raise ValueError(f"Unsupported file extension '{ext}'")
 
 def load_documents(source_dir: str, ignored_files: List[str] = []) -> List[Document]:
     """
@@ -100,7 +108,8 @@ def load_documents(source_dir: str, ignored_files: List[str] = []) -> List[Docum
         results = []
         with tqdm(total=len(filtered_files), desc='Loading new documents', ncols=80) as pbar:
             for i, docs in enumerate(pool.imap_unordered(load_single_document, filtered_files)):
-                results.extend(docs)
+                if docs:
+                    results.extend(docs)
                 pbar.update()
 
     return results

examples/langchain-python-rag-privategpt/requirements.txt

@@ -11,4 +11,5 @@ tabulate==0.9.0
 pandoc==2.3
 pypandoc==1.11
 tqdm==4.66.1
-sentence_transformers==2.2.2
+sentence_transformers==2.2.2
+numpy>=1.22.2 # not directly required, pinned by Snyk to avoid a vulnerability

format/format_test.go

@@ -5,7 +5,6 @@ import (
 )
 
 func TestHumanNumber(t *testing.T) {
-
 	type testCase struct {
 		input    uint64
 		expected string

go.mod

@@ -16,6 +16,7 @@ require (
 )
 
 require (
+	github.com/agnivade/levenshtein v1.1.1
 	github.com/d4l3k/go-bfloat16 v0.0.0-20211005043715-690c3bdd05f1
 	github.com/mattn/go-runewidth v0.0.14
 	github.com/nlpodyssey/gopickle v0.3.0

go.sum

@@ -4,10 +4,14 @@ dmitri.shuralyov.com/gpu/mtl v0.0.0-20190408044501-666a987793e9/go.mod h1:H6x//7
 gioui.org v0.0.0-20210308172011-57750fc8a0a6/go.mod h1:RSH6KIUZ0p2xy5zHDxgAM4zumjgTw83q2ge/PI+yyw8=
 github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03qcyfWMU=
 github.com/BurntSushi/xgb v0.0.0-20160522181843-27f122750802/go.mod h1:IVnqGOEym/WlBOVXweHU+Q+/VP0lqqI8lqeDx9IjBqo=
+github.com/agnivade/levenshtein v1.1.1 h1:QY8M92nrzkmr798gCo3kmMyqXFzdQVpxLlGPRBij0P8=
+github.com/agnivade/levenshtein v1.1.1/go.mod h1:veldBMzWxcCG2ZvUTKD2kJNRdCk5hVbJomOvKkmgYbo=
 github.com/ajstarks/svgo v0.0.0-20180226025133-644b8db467af/go.mod h1:K08gAheRH3/J6wwsYMMT4xOr94bZjxIelGM0+d/wbFw=
 github.com/antihax/optional v1.0.0/go.mod h1:uupD/76wgC+ih3iEmQUL+0Ugr19nfwCT1kdvxnR2qWY=
 github.com/apache/arrow/go/arrow v0.0.0-20211112161151-bc219186db40 h1:q4dksr6ICHXqG5hm0ZW5IHyeEJXoIJSOZeBLmWPNeIQ=
 github.com/apache/arrow/go/arrow v0.0.0-20211112161151-bc219186db40/go.mod h1:Q7yQnSMnLvcXlZ8RV+jwz/6y1rQTqbX6C82SndT52Zs=
+github.com/arbovm/levenshtein v0.0.0-20160628152529-48b4e1c0c4d0 h1:jfIu9sQUG6Ig+0+Ap1h4unLjW6YQJpKZVmUzxsD4E/Q=
+github.com/arbovm/levenshtein v0.0.0-20160628152529-48b4e1c0c4d0/go.mod h1:t2tdKJDJF9BV14lnkjHmOQgcvEKgtqs5a1N3LNdJhGE=
 github.com/boombuler/barcode v1.0.0/go.mod h1:paBWMcWSl3LHKBqUq+rly7CNSldXjb2rDl3JlRe0mD8=
 github.com/bytedance/sonic v1.11.6 h1:oUp34TzMlL+OY1OUWxHqsdkgC/Zfc85zGqw9siXjrc0=
 github.com/bytedance/sonic v1.11.6/go.mod h1:LysEHSvpvDySVdC2f87zGWf6CIKJcAvqab1ZaiQtds4=
@@ -36,6 +40,8 @@ github.com/d4l3k/go-bfloat16 v0.0.0-20211005043715-690c3bdd05f1/go.mod h1:uw2gLc
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/dgryski/trifles v0.0.0-20200323201526-dd97f9abfb48 h1:fRzb/w+pyskVMQ+UbP35JkH8yB7MYb4q/qhBarqZE6g=
+github.com/dgryski/trifles v0.0.0-20200323201526-dd97f9abfb48/go.mod h1:if7Fbed8SFyPtHLHbg49SI7NAdJiC5WIA09pe59rfAA=
 github.com/emirpasic/gods v1.18.1 h1:FXtiHYKDGKCW2KzwZKx0iC0PQmdlorYgdFG9jPXJ1Bc=
 github.com/emirpasic/gods v1.18.1/go.mod h1:8tpGGwCnJ5H4r6BWwaV6OrWmMoPhUl5jm/FMNAnJvWQ=
 github.com/envoyproxy/go-control-plane v0.9.0/go.mod h1:YTl/9mNaCwkRvm6d1a2C3ymFceY/DCBVvsKhRF0iEA4=

gpu/amd_windows.go

@@ -65,7 +65,7 @@ func AMDGetGPUInfo() []GpuInfo {
 
 	slog.Debug("detected hip devices", "count", count)
 	// TODO how to determine the underlying device ID when visible devices is causing this to subset?
-	for i := 0; i < count; i++ {
+	for i := range count {
 		err = hl.HipSetDevice(i)
 		if err != nil {
 			slog.Warn("set device", "id", i, "error", err)

gpu/assets.go

@@ -80,7 +80,7 @@ func cleanupTmpDirs() {
 		if err == nil {
 			pid, err := strconv.Atoi(string(raw))
 			if err == nil {
-				if proc, err := os.FindProcess(int(pid)); err == nil && !errors.Is(proc.Signal(syscall.Signal(0)), os.ErrProcessDone) {
+				if proc, err := os.FindProcess(pid); err == nil && !errors.Is(proc.Signal(syscall.Signal(0)), os.ErrProcessDone) {
 					// Another running ollama, ignore this tmpdir
 					continue
 				}

gpu/cuda_common.go

@@ -18,5 +18,4 @@ func cudaGetVisibleDevicesEnv(gpuInfo []GpuInfo) (string, string) {
 		ids = append(ids, info.ID)
 	}
 	return "CUDA_VISIBLE_DEVICES", strings.Join(ids, ",")
-
 }

gpu/gpu.go

@@ -16,13 +16,12 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
-	"strconv"
 	"strings"
 	"sync"
 	"unsafe"
 
-	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/envconfig"
+	"github.com/ollama/ollama/format"
 )
 
 type handles struct {
@@ -105,8 +104,6 @@ func initGPUHandles() *handles {
 	var cudartMgmtPatterns []string
 	var nvcudaMgmtName string
 	var nvcudaMgmtPatterns []string
-	var oneapiMgmtName string
-	var oneapiMgmtPatterns []string
 
 	tmpDir, _ := PayloadsDir()
 	switch runtime.GOOS {
@@ -118,8 +115,6 @@ func initGPUHandles() *handles {
 		// Aligned with driver, we can't carry as payloads
 		nvcudaMgmtName = "nvcuda.dll"
 		nvcudaMgmtPatterns = NvcudaWindowsGlobs
-		oneapiMgmtName = "ze_intel_gpu64.dll"
-		oneapiMgmtPatterns = OneapiWindowsGlobs
 	case "linux":
 		cudartMgmtName = "libcudart.so*"
 		if tmpDir != "" {
@@ -130,8 +125,6 @@ func initGPUHandles() *handles {
 		// Aligned with driver, we can't carry as payloads
 		nvcudaMgmtName = "libcuda.so*"
 		nvcudaMgmtPatterns = NvcudaLinuxGlobs
-		oneapiMgmtName = "libze_intel_gpu.so"
-		oneapiMgmtPatterns = OneapiLinuxGlobs
 	default:
 		return gpuHandles
 	}
@@ -159,17 +152,6 @@ func initGPUHandles() *handles {
 		}
 	}
 
-	oneapiLibPaths := FindGPULibs(oneapiMgmtName, oneapiMgmtPatterns)
-	if len(oneapiLibPaths) > 0 {
-		deviceCount, oneapi, libPath := LoadOneapiMgmt(oneapiLibPaths)
-		if oneapi != nil {
-			slog.Debug("detected Intel GPUs", "library", libPath, "count", deviceCount)
-			gpuHandles.oneapi = oneapi
-			gpuHandles.deviceCount = deviceCount
-			return gpuHandles
-		}
-	}
-
 	return gpuHandles
 }
 
@@ -205,7 +187,7 @@ func GetGPUInfo() GpuInfoList {
 	resp := []GpuInfo{}
 
 	// NVIDIA first
-	for i := 0; i < gpuHandles.deviceCount; i++ {
+	for i := range gpuHandles.deviceCount {
 		// TODO once we support CPU compilation variants of GPU libraries refine this...
 		if cpuVariant == "" && runtime.GOARCH == "amd64" {
 			continue
@@ -239,24 +221,12 @@ func GetGPUInfo() GpuInfoList {
 			gpuInfo.MinimumMemory = cudaMinimumMemory
 			gpuInfo.DependencyPath = depPath
 			gpuInfo.Name = C.GoString(&memInfo.gpu_name[0])
-			gpuInfo.DriverMajor = int(driverMajor)
-			gpuInfo.DriverMinor = int(driverMinor)
+			gpuInfo.DriverMajor = driverMajor
+			gpuInfo.DriverMinor = driverMinor
 
 			// TODO potentially sort on our own algorithm instead of what the underlying GPU library does...
 			resp = append(resp, gpuInfo)
 		}
-		if gpuHandles.oneapi != nil {
-			gpuInfo := GpuInfo{
-				Library: "oneapi",
-			}
-			C.oneapi_check_vram(*gpuHandles.oneapi, &memInfo)
-			var totalFreeMem float64 = float64(memInfo.free) * 0.95 // work-around: leave some reserve vram for mkl lib used in ggml-sycl backend.
-			memInfo.free = C.uint64_t(totalFreeMem)
-			gpuInfo.TotalMemory = uint64(memInfo.total)
-			gpuInfo.FreeMemory = uint64(memInfo.free)
-			gpuInfo.ID = strconv.Itoa(i)
-			resp = append(resp, gpuInfo)
-		}
 	}
 
 	// Then AMD
@@ -326,6 +296,7 @@ func FindGPULibs(baseLibName string, defaultPatterns []string) []string {
 		// Nvidia PhysX known to return bogus results
 		if strings.Contains(pattern, "PhysX") {
 			slog.Debug("skipping PhysX cuda library path", "path", pattern)
+			continue
 		}
 		// Ignore glob discovery errors
 		matches, _ := filepath.Glob(pattern)

gpu/gpu_test.go

@@ -5,11 +5,12 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestBasicGetGPUInfo(t *testing.T) {
 	info := GetGPUInfo()
-	assert.Greater(t, len(info), 0)
+	assert.NotEmpty(t, len(info))
 	assert.Contains(t, "cuda rocm cpu metal", info[0].Library)
 	if info[0].Library != "cpu" {
 		assert.Greater(t, info[0].TotalMemory, uint64(0))
@@ -19,7 +20,7 @@ func TestBasicGetGPUInfo(t *testing.T) {
 
 func TestCPUMemInfo(t *testing.T) {
 	info, err := GetCPUMem()
-	assert.NoError(t, err)
+	require.NoError(t, err)
 	switch runtime.GOOS {
 	case "darwin":
 		t.Skip("CPU memory not populated on darwin")

llm/ext_server/server.cpp

@@ -140,7 +140,6 @@ struct server_slot {
     std::vector<llama_token> cache_tokens;
     std::vector<completion_token_output> generated_token_probs;
 
-    bool infill = false;
     bool embedding = false;
     bool has_next_token = true;
     bool truncated = false;
@@ -187,7 +186,6 @@ struct server_slot {
         n_past                 = 0;
         n_sent_text            = 0;
         n_sent_token_probs     = 0;
-        infill                 = false;
         ga_i                   = 0;
         n_past_se              = 0;
 
@@ -361,7 +359,6 @@ struct llama_server_context
 
     // slots / clients
     std::vector<server_slot> slots;
-    json default_generation_settings_for_props;
 
     llama_server_queue    queue_tasks;
     llama_server_response queue_results;
@@ -485,9 +482,6 @@ struct llama_server_context
             slots.push_back(slot);
         }
 
-        default_generation_settings_for_props = get_formated_generation(slots.front());
-        default_generation_settings_for_props["seed"] = -1;
-
         batch = llama_batch_init(n_ctx, 0, params.n_parallel);
     }
 
@@ -586,7 +580,7 @@ struct llama_server_context
         slot->sparams.mirostat_eta      = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
         slot->sparams.penalize_nl       = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
         slot->params.n_keep             = json_value(data, "n_keep",            slot->params.n_keep);
-        slot->params.seed               = json_value(data, "seed",              default_params.seed);
+        slot->sparams.seed              = json_value(data, "seed",              default_params.seed);
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
@@ -600,16 +594,6 @@ struct llama_server_context
             slot->params.n_predict = slot->n_predict;
         }
 
-        // infill
-        if (data.count("input_prefix") != 0)
-        {
-            slot->params.input_prefix = data["input_prefix"];
-        }
-        else
-        {
-            slot->params.input_prefix = "";
-        }
-
         if (data.count("input_suffix") != 0)
         {
             slot->params.input_suffix = data["input_suffix"];
@@ -823,7 +807,6 @@ struct llama_server_context
             llama_sampling_free(slot->ctx_sampling);
         }
         slot->ctx_sampling = llama_sampling_init(slot->sparams);
-        llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
         all_slots_are_idle = false;
@@ -847,7 +830,7 @@ struct llama_server_context
         system_tokens.clear();
 
         if (!system_prompt.empty()) {
-            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = ::llama_tokenize(ctx, system_prompt, true);
 
             llama_batch_clear(batch);
 
@@ -897,15 +880,6 @@ struct llama_server_context
         system_need_update = true;
     }
 
-    void system_prompt_process(const json &sys_props) {
-        system_prompt  = sys_props.value("prompt", "");
-        name_user      = sys_props.value("anti_prompt", "");
-        name_assistant = sys_props.value("assistant_name", "");
-
-
-        system_prompt_notify();
-    }
-
     static size_t find_stopping_strings(const std::string &text, const size_t last_token_size,
                                         const stop_type type, server_slot &slot)
     {
@@ -1263,13 +1237,12 @@ struct llama_server_context
         queue_results.send(res);
     }
 
-    void request_completion(int task_id, json data, bool infill, bool embedding, int multitask_id)
+    void request_completion(int task_id, json data, bool embedding, int multitask_id)
     {
         task_server task;
         task.id = task_id;
         task.target_id = 0;
         task.data = std::move(data);
-        task.infill_mode = infill;
         task.embedding_mode = embedding;
         task.type = TASK_TYPE_COMPLETION;
         task.multitask_id = multitask_id;
@@ -1415,8 +1388,8 @@ struct llama_server_context
             json subtask_data = multiprompt_task.data;
             subtask_data["prompt"] = subtask_data["prompt"][i];
 
-            // subtasks inherit everything else (infill mode, embedding mode, etc.)
-            request_completion(subtask_ids[i], subtask_data, multiprompt_task.infill_mode, multiprompt_task.embedding_mode, multitask_id);
+            // subtasks inherit everything else (embedding mode, etc.)
+            request_completion(subtask_ids[i], subtask_data, multiprompt_task.embedding_mode, multitask_id);
         }
     }
 
@@ -1434,26 +1407,8 @@ struct llama_server_context
                     break;
                 }
 
-                if (task.data.contains("system_prompt"))
-                {
-                    if (!all_slots_are_idle) {
-                        send_error(task, "system prompt can only be updated when all slots are idle");
-                        break;
-                    }
-                    system_prompt_process(task.data["system_prompt"]);
-
-                    // reset cache_tokens for all slots
-                    for (server_slot &slot : slots)
-                    {
-                        slot.cache_tokens.clear();
-                        slot.n_past    = 0;
-                        slot.n_past_se = 0;
-                    }
-                }
-
                 slot->reset();
 
-                slot->infill       = task.infill_mode;
                 slot->embedding    = task.embedding_mode;
                 slot->task_id      = task.id;
                 slot->multitask_id = task.multitask_id;
@@ -1679,8 +1634,7 @@ struct llama_server_context
                 const bool has_prompt = slot.prompt.is_array() || (slot.prompt.is_string() && !slot.prompt.get<std::string>().empty()) || !slot.images.empty();
 
                 // empty prompt passed -> release the slot and send empty response
-                // note: infill mode allows empty prompt
-                if (slot.state == IDLE && slot.command == LOAD_PROMPT && !has_prompt && !slot.infill)
+                if (slot.state == IDLE && slot.command == LOAD_PROMPT && !has_prompt)
                 {
                     slot.release();
                     slot.print_timings();
@@ -1697,33 +1651,7 @@ struct llama_server_context
                     slot.t_start_process_prompt = ggml_time_us();
                     slot.t_start_genereration = 0;
 
-                    if (slot.infill)
-                    {
-                        bool suff_rm_leading_spc = true;
-                        if (params.input_suffix.find_first_of(' ') == 0 && params.input_suffix.size() > 1)
-                        {
-                            params.input_suffix.erase(0, 1);
-                            suff_rm_leading_spc = false;
-                        }
-                        auto prefix_tokens = tokenize(slot.params.input_prefix, false);
-                        auto suffix_tokens = tokenize(slot.params.input_suffix, false);
-
-                        const int space_token = 29871; // TODO: this should not be hardcoded
-                        if (suff_rm_leading_spc && !suffix_tokens.empty() && suffix_tokens[0] == space_token) {
-                            suffix_tokens.erase(suffix_tokens.begin());
-                        }
-
-                        prefix_tokens.insert(prefix_tokens.begin(), llama_token_prefix(model));
-                        prefix_tokens.insert(prefix_tokens.begin(), llama_token_bos(model)); // always add BOS
-                        prefix_tokens.insert(prefix_tokens.end(),   llama_token_suffix(model));
-                        prefix_tokens.insert(prefix_tokens.end(),   suffix_tokens.begin(), suffix_tokens.end());
-                        prefix_tokens.push_back(llama_token_middle(model));
-                        prompt_tokens = prefix_tokens;
-                    }
-                    else
-                    {
-                        prompt_tokens = tokenize(slot.prompt, system_prompt.empty() && add_bos_token);  // add BOS if there isn't system prompt
-                    }
+                    prompt_tokens = tokenize(slot.prompt, system_prompt.empty());  // add BOS if there isn't system prompt
 
                     slot.n_prompt_tokens = prompt_tokens.size();
 
@@ -2130,8 +2058,7 @@ static void server_print_usage(const char *argv0, const gpt_params &params,
     printf("\n");
 }
 
-static void server_params_parse(int argc, char **argv, server_params &sparams,
-                                gpt_params &params, llama_server_context& llama)
+static void server_params_parse(int argc, char **argv, server_params &sparams, gpt_params &params)
 {
     gpt_params default_params;
     server_params default_sparams;
@@ -2546,27 +2473,6 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
             }
             params.n_predict = std::stoi(argv[i]);
         }
-        else if (arg == "-spf" || arg == "--system-prompt-file")
-        {
-            if (++i >= argc)
-            {
-                invalid_param = true;
-                break;
-            }
-            std::ifstream file(argv[i]);
-            if (!file) {
-                fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-                invalid_param = true;
-                break;
-            }
-            std::string systm_content;
-            std::copy(
-                std::istreambuf_iterator<char>(file),
-                std::istreambuf_iterator<char>(),
-                std::back_inserter(systm_content)
-            );
-            llama.system_prompt_process(json::parse(systm_content));
-        }
         else if (arg == "-ctk" || arg == "--cache-type-k") {
             params.cache_type_k = argv[++i];
         }
@@ -2818,7 +2724,7 @@ int main(int argc, char **argv) {
     // struct that contains llama context and inference
     llama_server_context llama;
 
-    server_params_parse(argc, argv, sparams, params, llama);
+    server_params_parse(argc, argv, sparams, params);
 
     if (params.model_alias == "unknown")
     {
@@ -3150,7 +3056,7 @@ int main(int argc, char **argv) {
                 json data = json::parse(req.body);
                 const int task_id = llama.queue_tasks.get_new_id();
                 llama.queue_results.add_waiting_task_id(task_id);
-                llama.request_completion(task_id, data, false, false, -1);
+                llama.request_completion(task_id, data, false, -1);
                 if (!json_value(data, "stream", false)) {
                     std::string completion_text;
                     task_result result = llama.queue_results.recv(task_id);
@@ -3272,7 +3178,7 @@ int main(int argc, char **argv) {
                 // create and queue the task
                 const int task_id = llama.queue_tasks.get_new_id();
                 llama.queue_results.add_waiting_task_id(task_id);
-                llama.request_completion(task_id, { {"prompt", prompt}, { "n_predict", 0}, {"image_data", image_data} }, false, true, -1);
+                llama.request_completion(task_id, { {"prompt", prompt}, { "n_predict", 0}, {"image_data", image_data} }, true, -1);
 
                 // get the result
                 task_result result = llama.queue_results.recv(task_id);

llm/generate/gen_darwin.sh

@@ -32,42 +32,43 @@ case "${GOARCH}" in
     echo "Building static library"
     build
 
+    if [ -z "$OLLAMA_SKIP_CPU_GENERATE" ]; then
+        #
+        # CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
+        #
+        init_vars
+        CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=off -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
+        BUILD_DIR="../build/darwin/${ARCH}/cpu"
+        echo "Building LCD CPU"
+        build
+        sign ${BUILD_DIR}/bin/ollama_llama_server
+        compress
 
-    #
-    # CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
-    #
-    init_vars
-    CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=off -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
-    BUILD_DIR="../build/darwin/${ARCH}/cpu"
-    echo "Building LCD CPU"
-    build
-    sign ${BUILD_DIR}/bin/ollama_llama_server
-    compress
+        #
+        # ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
+        # Approximately 400% faster than LCD on same CPU
+        #
+        init_vars
+        CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
+        BUILD_DIR="../build/darwin/${ARCH}/cpu_avx"
+        echo "Building AVX CPU"
+        build
+        sign ${BUILD_DIR}/bin/ollama_llama_server
+        compress
 
-    #
-    # ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
-    # Approximately 400% faster than LCD on same CPU
-    #
-    init_vars
-    CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
-    BUILD_DIR="../build/darwin/${ARCH}/cpu_avx"
-    echo "Building AVX CPU"
-    build
-    sign ${BUILD_DIR}/bin/ollama_llama_server
-    compress
-
-    #
-    # ~2013 CPU Dynamic library
-    # Approximately 10% faster than AVX on same CPU
-    #
-    init_vars
-    CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=on -DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_AVX512=off -DLLAMA_FMA=on -DLLAMA_F16C=on ${CMAKE_DEFS}"
-    BUILD_DIR="../build/darwin/${ARCH}/cpu_avx2"
-    echo "Building AVX2 CPU"
-    EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation"
-    build
-    sign ${BUILD_DIR}/bin/ollama_llama_server
-    compress
+        #
+        # ~2013 CPU Dynamic library
+        # Approximately 10% faster than AVX on same CPU
+        #
+        init_vars
+        CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_ACCELERATE=on -DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_AVX512=off -DLLAMA_FMA=on -DLLAMA_F16C=on ${CMAKE_DEFS}"
+        BUILD_DIR="../build/darwin/${ARCH}/cpu_avx2"
+        echo "Building AVX2 CPU"
+        EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation"
+        build
+        sign ${BUILD_DIR}/bin/ollama_llama_server
+        compress
+    fi
     ;;
 "arm64")
 
@@ -79,13 +80,15 @@ case "${GOARCH}" in
     echo "Building static library"
     build
 
-    init_vars
-    CMAKE_DEFS="${COMMON_DARWIN_DEFS} -DLLAMA_ACCELERATE=on -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} -DLLAMA_METAL=on ${CMAKE_DEFS}"
-    BUILD_DIR="../build/darwin/${ARCH}/metal"
-    EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders"
-    build
-    sign ${BUILD_DIR}/bin/ollama_llama_server
-    compress
+    if [ -z "$OLLAMA_SKIP_METAL_GENERATE" ]; then
+        init_vars
+        CMAKE_DEFS="${COMMON_DARWIN_DEFS} -DLLAMA_ACCELERATE=on -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} -DLLAMA_METAL=on ${CMAKE_DEFS}"
+        BUILD_DIR="../build/darwin/${ARCH}/metal"
+        EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders"
+        build
+        sign ${BUILD_DIR}/bin/ollama_llama_server
+        compress
+    fi
     ;;
 *)
     echo "GOARCH must be set"

llm/generate/gen_linux.sh

@@ -211,7 +211,7 @@ if [ -z "${ONEAPI_ROOT}" ]; then
     ONEAPI_ROOT=/opt/intel/oneapi
 fi
 
-if [ -d "${ONEAPI_ROOT}" ]; then
+if [ -z "${OLLAMA_SKIP_ONEAPI_GENERATE}" -a -d "${ONEAPI_ROOT}" ]; then
     echo "OneAPI libraries detected - building dynamic OneAPI library"
     init_vars
     source ${ONEAPI_ROOT}/setvars.sh --force # set up environment variables for oneAPI

llm/generate/gen_windows.ps1

@@ -290,7 +290,7 @@ function build_cuda() {
 }
 
 function build_oneapi() {
-  if ((-not "${env:OLLAMA_SKIP_CUDA_GENERATE}") -and ("${env:ONEAPI_ROOT}"))  {
+  if ((-not "${env:OLLAMA_SKIP_ONEAPI_GENERATE}") -and ("${env:ONEAPI_ROOT}"))  {
     # Get oneAPI version
     $script:ONEAPI_VERSION = icpx --version
     $script:ONEAPI_VERSION = [regex]::Match($script:ONEAPI_VERSION, '(?<=oneAPI DPC\+\+/C\+\+ Compiler )(?<version>\d+\.\d+\.\d+)').Value

llm/ggml.go

@@ -81,6 +81,11 @@ func (kv KV) ContextLength() uint64 {
 	return kv.u64(fmt.Sprintf("%s.context_length", kv.Architecture()))
 }
 
+func (kv KV) ChatTemplate() string {
+	s, _ := kv["tokenizer.chat_template"].(string)
+	return s
+}
+
 type Tensors []*Tensor
 
 func (ts Tensors) Layers() map[string]Layer {

llm/gguf.go

@@ -592,8 +592,8 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
 			return err
 		}
 
-		dims := 0
-		for cnt := 0; cnt < len(tensor.Shape); cnt++ {
+		var dims int
+		for cnt := range len(tensor.Shape) {
 			if tensor.Shape[cnt] > 0 {
 				dims++
 			}
@@ -603,8 +603,8 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
 			return err
 		}
 
-		for i := 0; i < dims; i++ {
-			if err := binary.Write(ws, llm.ByteOrder, uint64(tensor.Shape[dims-1-i])); err != nil {
+		for i := range dims {
+			if err := binary.Write(ws, llm.ByteOrder, tensor.Shape[dims-1-i]); err != nil {
 				return err
 			}
 		}
@@ -618,22 +618,8 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
 		}
 	}
 
-	offset, err := ws.Seek(0, io.SeekCurrent)
-	if err != nil {
-		return err
-	}
-
 	var alignment int64 = 32
-	padding := llm.padding(offset, alignment)
-	if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding))); err != nil {
-		return err
-	}
-
 	for _, tensor := range tensors {
-		if _, err := tensor.WriteTo(ws); err != nil {
-			return err
-		}
-
 		offset, err := ws.Seek(0, io.SeekCurrent)
 		if err != nil {
 			return err
@@ -643,6 +629,10 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
 		if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding))); err != nil {
 			return err
 		}
+
+		if _, err := tensor.WriteTo(ws); err != nil {
+			return err
+		}
 	}
 
 	return nil

llm/memory.go

@@ -5,9 +5,9 @@ import (
 	"log/slog"
 
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/gpu"
-	"github.com/ollama/ollama/envconfig"
 )
 
 // This algorithm looks for a complete fit to determine if we need to unload other models
@@ -103,7 +103,7 @@ func EstimateGPULayers(gpus []gpu.GpuInfo, ggml *GGML, projectors []string, opts
 	}
 
 	var layerCount int
-	for i := 0; i < int(ggml.KV().BlockCount()); i++ {
+	for i := range int(ggml.KV().BlockCount()) {
 		if blk, ok := layers[fmt.Sprintf("blk.%d", i)]; ok {
 			memoryLayer := blk.size()
 

llm/patches/05-default-pretokenizer.diff

@@ -1,35 +1,32 @@
-From d02a06f3f45a09255ace8684a66590e06ce44605 Mon Sep 17 00:00:00 2001
-From: Michael Yang <mxyng@pm.me>
-Date: Thu, 23 May 2024 11:33:20 -0700
-Subject: [PATCH] default pretokenizer on unrecognized type
-
----
- llama.cpp | 5 +----
- 1 file changed, 1 insertion(+), 4 deletions(-)
-
 diff --git a/llama.cpp b/llama.cpp
-index 15c66077..af1aede3 100644
+index 40d2ec2c..74f3ee9c 100644
 --- a/llama.cpp
 +++ b/llama.cpp
-@@ -4504,9 +4504,6 @@ static void llm_load_vocab(
-                 LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
-                 LLAMA_LOG_WARN("%s:                                             \n", __func__);
-                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
--            } else if (
--                    tokenizer_pre == "default") {
+@@ -4642,16 +4642,7 @@ static void llm_load_vocab(
+ 
+         // for now, only BPE models have pre-tokenizers
+         if (vocab.type == LLAMA_VOCAB_TYPE_BPE) {
+-            if (tokenizer_pre.empty()) {
+-                LLAMA_LOG_WARN("%s: missing pre-tokenizer type, using: 'default'\n", __func__);
+-                LLAMA_LOG_WARN("%s:                                             \n", __func__);
+-                LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
+-                LLAMA_LOG_WARN("%s: GENERATION QUALITY WILL BE DEGRADED!        \n", __func__);
+-                LLAMA_LOG_WARN("%s: CONSIDER REGENERATING THE MODEL             \n", __func__);
+-                LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
+-                LLAMA_LOG_WARN("%s:                                             \n", __func__);
 -                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+-            } else if (
++            if (
+                     tokenizer_pre == "default") {
+                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
              } else if (
-                     tokenizer_pre == "llama3"   ||
-                     tokenizer_pre == "llama-v3" ||
-@@ -4553,7 +4550,7 @@ static void llm_load_vocab(
-                 tokenizer_pre == "dbrx") {
-                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DBRX;
+@@ -4703,7 +4694,8 @@ static void llm_load_vocab(
+                 tokenizer_pre == "smaug-bpe") {
+                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMAUG;
              } else {
 -                throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
++                LLAMA_LOG_WARN("%s: missing or unrecognized pre-tokenizer type, using: 'default'\n", __func__);
 +                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
              }
          } else {
              vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
--- 
-2.45.1
-

llm/patches/06-qwen2.diff

@@ -0,0 +1,13 @@
+diff --git a/llama.cpp b/llama.cpp
+index 40d2ec2c..f34eb79a 100644
+--- a/llama.cpp
++++ b/llama.cpp
+@@ -6943,7 +6943,7 @@ static struct ggml_tensor * llm_build_kqv(
+         struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
+         cb(kq, "kq", il);
+ 
+-        if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) {
++        if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX || model.arch == LLM_ARCH_QWEN2) {
+             // for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs
+             // ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847
+             ggml_mul_mat_set_prec(kq, GGML_PREC_F32);

llm/payload.go

@@ -10,9 +10,9 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"slices"
 	"strings"
 
-	"golang.org/x/exp/slices"
 	"golang.org/x/sync/errgroup"
 
 	"github.com/ollama/ollama/gpu"

llm/server.go

@@ -85,7 +85,6 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 	var systemMemory uint64
 	gpuCount := len(gpus)
 	if (len(gpus) == 1 && gpus[0].Library == "cpu") || opts.NumGPU == 0 {
-
 		// TODO evaluate system memory to see if we should block the load, or force an unload of another CPU runner
 
 		cpuRunner = serverForCpu()
@@ -104,21 +103,22 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 		var layers int
 		layers, estimatedVRAM, estimatedTotal = EstimateGPULayers(gpus, ggml, projectors, opts)
 
-		if gpus[0].Library == "metal" && estimatedVRAM > systemMemory {
+		switch {
+		case gpus[0].Library == "metal" && estimatedVRAM > systemMemory:
 			// disable partial offloading when model is greater than total system memory as this
 			// can lead to locking up the system
 			opts.NumGPU = 0
-		} else if gpus[0].Library != "metal" && layers == 0 {
+		case gpus[0].Library != "metal" && layers == 0:
 			// Don't bother loading into the GPU if no layers can fit
 			cpuRunner = serverForCpu()
 			gpuCount = 0
-		} else if opts.NumGPU < 0 && layers > 0 && gpus[0].Library != "cpu" {
+		case opts.NumGPU < 0 && layers > 0 && gpus[0].Library != "cpu":
 			opts.NumGPU = layers
 		}
 	}
 
 	// Loop through potential servers
-	finalErr := fmt.Errorf("no suitable llama servers found")
+	finalErr := errors.New("no suitable llama servers found")
 
 	if len(adapters) > 1 {
 		return nil, errors.New("ollama supports only one lora adapter, but multiple were provided")
@@ -189,35 +189,38 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 		params = append(params, "--memory-f32")
 	}
 
-	if opts.UseMLock {
-		params = append(params, "--mlock")
+	flashAttnEnabled := envconfig.FlashAttention
+
+	for _, g := range gpus {
+		// only cuda (compute capability 7+) and metal support flash attention
+		if g.Library != "metal" && (g.Library != "cuda" || g.DriverMajor < 7) {
+			flashAttnEnabled = false
+		}
+
+		// mmap has issues with partial offloading on metal
+		if g.Library == "metal" &&
+			uint64(opts.NumGPU) > 0 &&
+			uint64(opts.NumGPU) < ggml.KV().BlockCount()+1 {
+			opts.UseMMap = false
+		}
+	}
+
+	if flashAttnEnabled {
+		params = append(params, "--flash-attn")
 	}
 
 	if !opts.UseMMap {
 		params = append(params, "--no-mmap")
 	}
 
+	if opts.UseMLock {
+		params = append(params, "--mlock")
+	}
+
 	if opts.UseNUMA {
 		params = append(params, "--numa")
 	}
 
-	flashAttnEnabled := envconfig.FlashAttention
-
-	// partial offloading does not support flash attention
-	if uint64(opts.NumGPU) < ggml.KV().BlockCount()+1 {
-		flashAttnEnabled = false
-	}
-
-	// only cuda (compute capability 7+) and metal support flash attention
-	for _, g := range gpus {
-		if g.Library != "metal" && (g.Library != "cuda" || g.DriverMajor < 7) {
-			flashAttnEnabled = false
-		}
-	}
-	if flashAttnEnabled {
-		params = append(params, "--flash-attn")
-	}
-
 	numParallel := envconfig.NumParallel
 
 	// TODO (jmorganca): multimodal models don't support parallel yet
@@ -229,7 +232,7 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 
 	params = append(params, "--parallel", fmt.Sprintf("%d", numParallel))
 
-	for i := 0; i < len(servers); i++ {
+	for i := range len(servers) {
 		dir := availableServers[servers[i]]
 		if dir == "" {
 			// Shouldn't happen
@@ -281,7 +284,7 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 
 		server := filepath.Join(dir, "ollama_llama_server")
 		if runtime.GOOS == "windows" {
-			server = server + ".exe"
+			server += ".exe"
 		}
 
 		// Detect tmp cleaners wiping out the file
@@ -312,7 +315,7 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 		s.cmd.Stdout = os.Stdout
 		s.cmd.Stderr = s.status
 
-		visibleDevicesEnv, visibleDevicesEnvVal := gpu.GpuInfoList(gpus).GetVisibleDevicesEnv()
+		visibleDevicesEnv, visibleDevicesEnvVal := gpus.GetVisibleDevicesEnv()
 		pathEnvVal := strings.Join(libraryPaths, string(filepath.ListSeparator))
 
 		// Update or add the path and visible devices variable with our adjusted version
@@ -456,7 +459,7 @@ func (s *llmServer) getServerStatus(ctx context.Context) (ServerStatus, error) {
 	resp, err := http.DefaultClient.Do(req)
 	if err != nil {
 		if errors.Is(err, context.DeadlineExceeded) {
-			return ServerStatusNotResponding, fmt.Errorf("server not responding")
+			return ServerStatusNotResponding, errors.New("server not responding")
 		}
 		return ServerStatusError, fmt.Errorf("health resp: %w", err)
 	}
@@ -603,7 +606,7 @@ array  ::=
 
 string ::=
   "\"" (
-    [^"\\] |
+    [^"\\\x7F\x00-\x1F] |
     "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
   )* "\"" ws
 

openai/openai.go

@@ -211,6 +211,29 @@ func fromRequest(r ChatCompletionRequest) api.ChatRequest {
 	}
 }
 
+type DeleteCompletion struct {
+	Id      string `json:"id"`
+	Object  string `json:"object"`
+	Deleted bool   `json:"deleted"`
+}
+
+func toDeleteCompletion(model string) DeleteCompletion {
+	return DeleteCompletion{
+		Id:      model,
+		Object:  "model",
+		Deleted: true,
+	}
+}
+
+type BaseWriter struct {
+	gin.ResponseWriter
+}
+
+type DeleteWriter struct {
+	BaseWriter
+	model string
+}
+
 type writer struct {
 	stream bool
 	id     string
@@ -245,7 +268,6 @@ func (w *writer) writeResponse(data []byte) (int, error) {
 		d, err := json.Marshal(toChunk(w.id, chatResponse))
 		if err != nil {
 			return 0, err
-
 		}
 
 		w.ResponseWriter.Header().Set("Content-Type", "text/event-stream")
@@ -316,3 +338,69 @@ func Middleware() gin.HandlerFunc {
 		c.Next()
 	}
 }
+
+func (w *DeleteWriter) writeError(code int, data []byte) (int, error) {
+	var serr api.StatusError
+	err := json.Unmarshal(data, &serr)
+	if err != nil {
+		return 0, err
+	}
+
+	w.ResponseWriter.Header().Set("Content-Type", "application/json")
+	err = json.NewEncoder(w.ResponseWriter).Encode(NewError(http.StatusInternalServerError, serr.Error()))
+	if err != nil {
+		return 0, err
+	}
+
+	return len(data), nil
+}
+
+func (w *DeleteWriter) writeResponse(data []byte) (int, error) {
+	// delete completion
+	w.ResponseWriter.Header().Set("Content-Type", "application/json")
+	err := json.NewEncoder(w.ResponseWriter).Encode(toDeleteCompletion(w.model))
+	if err != nil {
+		return 0, err
+	}
+
+	return len(data), nil
+}
+
+func (w *DeleteWriter) Write(data []byte) (int, error) {
+	code := w.ResponseWriter.Status()
+	if code != http.StatusOK {
+		return w.writeError(code, data)
+	}
+
+	return w.writeResponse(data)
+}
+
+func DeleteMiddleware() gin.HandlerFunc {
+	return func(c *gin.Context) {
+		var b bytes.Buffer
+		if err := json.NewEncoder(&b).Encode(api.DeleteRequest{Model: c.Param("model")}); err != nil {
+			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
+			return
+		}
+
+		c.Request.Body = io.NopCloser(&b)
+
+		// response writer
+		w := &DeleteWriter{
+			BaseWriter: BaseWriter{ResponseWriter: c.Writer},
+			model:      c.Param("model"),
+		}
+
+		c.Writer = w
+
+		c.Next()
+
+		// If the status code is OK, write the DeleteCompletion response
+		if c.Writer.Status() == http.StatusOK {
+			_, err := w.writeResponse(nil)
+			if err != nil {
+				c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
+			}
+		}
+	}
+}

parser/parser.go

@@ -8,7 +8,9 @@ import (
 	"io"
 	"strconv"
 	"strings"
-	"unicode"
+
+	"golang.org/x/text/encoding/unicode"
+	"golang.org/x/text/transform"
 )
 
 type File struct {
@@ -69,14 +71,11 @@ func ParseFile(r io.Reader) (*File, error) {
 	var b bytes.Buffer
 	var role string
 
-	var lineCount int
-	var linePos int
-
-	var utf16 bool
-
 	var f File
 
-	br := bufio.NewReader(r)
+	tr := unicode.BOMOverride(unicode.UTF8.NewDecoder())
+	br := bufio.NewReader(transform.NewReader(r, tr))
+
 	for {
 		r, _, err := br.ReadRune()
 		if errors.Is(err, io.EOF) {
@@ -85,17 +84,6 @@ func ParseFile(r io.Reader) (*File, error) {
 			return nil, err
 		}
 
-		// the utf16 byte order mark will be read as "unreadable" by ReadRune()
-		if isUnreadable(r) && lineCount == 0 && linePos == 0 {
-			utf16 = true
-			continue
-		}
-
-		// skip the second byte if we're reading utf16
-		if utf16 && r == 0 {
-			continue
-		}
-
 		next, r, err := parseRuneForState(r, curr)
 		if errors.Is(err, io.ErrUnexpectedEOF) {
 			return nil, fmt.Errorf("%w: %s", err, b.String())
@@ -103,13 +91,6 @@ func ParseFile(r io.Reader) (*File, error) {
 			return nil, err
 		}
 
-		if isNewline(r) {
-			lineCount++
-			linePos = 0
-		} else {
-			linePos++
-		}
-
 		// process the state transition, some transitions need to be intercepted and redirected
 		if next != curr {
 			switch curr {
@@ -309,10 +290,6 @@ func isNewline(r rune) bool {
 	return r == '\r' || r == '\n'
 }
 
-func isUnreadable(r rune) bool {
-	return r == unicode.ReplacementChar
-}
-
 func isValidMessageRole(role string) bool {
 	return role == "system" || role == "user" || role == "assistant"
 }

parser/parser_test.go

@@ -10,6 +10,9 @@ import (
 	"unicode/utf16"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/text/encoding"
+	"golang.org/x/text/encoding/unicode"
 )
 
 func TestParseFileFile(t *testing.T) {
@@ -25,7 +28,7 @@ TEMPLATE template1
 	reader := strings.NewReader(input)
 
 	modelfile, err := ParseFile(reader)
-	assert.NoError(t, err)
+	require.NoError(t, err)
 
 	expectedCommands := []Command{
 		{Name: "model", Args: "model1"},
@@ -88,7 +91,7 @@ func TestParseFileFrom(t *testing.T) {
 	for _, c := range cases {
 		t.Run("", func(t *testing.T) {
 			modelfile, err := ParseFile(strings.NewReader(c.input))
-			assert.ErrorIs(t, err, c.err)
+			require.ErrorIs(t, err, c.err)
 			if modelfile != nil {
 				assert.Equal(t, c.expected, modelfile.Commands)
 			}
@@ -105,7 +108,7 @@ PARAMETER param1
 	reader := strings.NewReader(input)
 
 	_, err := ParseFile(reader)
-	assert.ErrorIs(t, err, io.ErrUnexpectedEOF)
+	require.ErrorIs(t, err, io.ErrUnexpectedEOF)
 }
 
 func TestParseFileBadCommand(t *testing.T) {
@@ -114,8 +117,7 @@ FROM foo
 BADCOMMAND param1 value1
 `
 	_, err := ParseFile(strings.NewReader(input))
-	assert.ErrorIs(t, err, errInvalidCommand)
-
+	require.ErrorIs(t, err, errInvalidCommand)
 }
 
 func TestParseFileMessages(t *testing.T) {
@@ -201,7 +203,7 @@ MESSAGE system`,
 	for _, c := range cases {
 		t.Run("", func(t *testing.T) {
 			modelfile, err := ParseFile(strings.NewReader(c.input))
-			assert.ErrorIs(t, err, c.err)
+			require.ErrorIs(t, err, c.err)
 			if modelfile != nil {
 				assert.Equal(t, c.expected, modelfile.Commands)
 			}
@@ -355,7 +357,7 @@ TEMPLATE """
 	for _, c := range cases {
 		t.Run("", func(t *testing.T) {
 			modelfile, err := ParseFile(strings.NewReader(c.multiline))
-			assert.ErrorIs(t, err, c.err)
+			require.ErrorIs(t, err, c.err)
 			if modelfile != nil {
 				assert.Equal(t, c.expected, modelfile.Commands)
 			}
@@ -413,7 +415,7 @@ func TestParseFileParameters(t *testing.T) {
 			fmt.Fprintln(&b, "FROM foo")
 			fmt.Fprintln(&b, "PARAMETER", k)
 			modelfile, err := ParseFile(&b)
-			assert.NoError(t, err)
+			require.NoError(t, err)
 
 			assert.Equal(t, []Command{
 				{Name: "model", Args: "foo"},
@@ -442,7 +444,7 @@ FROM foo
 	for _, c := range cases {
 		t.Run("", func(t *testing.T) {
 			modelfile, err := ParseFile(strings.NewReader(c.input))
-			assert.NoError(t, err)
+			require.NoError(t, err)
 			assert.Equal(t, c.expected, modelfile.Commands)
 		})
 	}
@@ -501,15 +503,14 @@ SYSTEM ""
 	for _, c := range cases {
 		t.Run("", func(t *testing.T) {
 			modelfile, err := ParseFile(strings.NewReader(c))
-			assert.NoError(t, err)
+			require.NoError(t, err)
 
 			modelfile2, err := ParseFile(strings.NewReader(modelfile.String()))
-			assert.NoError(t, err)
+			require.NoError(t, err)
 
 			assert.Equal(t, modelfile, modelfile2)
 		})
 	}
-
 }
 
 func TestParseFileUTF16ParseFile(t *testing.T) {
@@ -518,14 +519,6 @@ PARAMETER param1 1
 PARAMETER param2 4096
 SYSTEM You are a utf16 file.
 `
-	// simulate a utf16 le file
-	utf16File := utf16.Encode(append([]rune{'\ufffe'}, []rune(data)...))
-	buf := new(bytes.Buffer)
-	err := binary.Write(buf, binary.LittleEndian, utf16File)
-	assert.NoError(t, err)
-
-	actual, err := ParseFile(buf)
-	assert.NoError(t, err)
 
 	expected := []Command{
 		{Name: "model", Args: "bob"},
@@ -534,14 +527,52 @@ SYSTEM You are a utf16 file.
 		{Name: "system", Args: "You are a utf16 file."},
 	}
 
-	assert.Equal(t, expected, actual.Commands)
+	t.Run("le", func(t *testing.T) {
+		var b bytes.Buffer
+		require.NoError(t, binary.Write(&b, binary.LittleEndian, []byte{0xff, 0xfe}))
+		require.NoError(t, binary.Write(&b, binary.LittleEndian, utf16.Encode([]rune(data))))
 
-	// simulate a utf16 be file
-	buf = new(bytes.Buffer)
-	err = binary.Write(buf, binary.BigEndian, utf16File)
-	assert.NoError(t, err)
+		actual, err := ParseFile(&b)
+		require.NoError(t, err)
 
-	actual, err = ParseFile(buf)
-	assert.NoError(t, err)
-	assert.Equal(t, expected, actual.Commands)
+		assert.Equal(t, expected, actual.Commands)
+	})
+
+	t.Run("be", func(t *testing.T) {
+		var b bytes.Buffer
+		require.NoError(t, binary.Write(&b, binary.BigEndian, []byte{0xfe, 0xff}))
+		require.NoError(t, binary.Write(&b, binary.BigEndian, utf16.Encode([]rune(data))))
+
+		actual, err := ParseFile(&b)
+		require.NoError(t, err)
+		assert.Equal(t, expected, actual.Commands)
+	})
+}
+
+func TestParseMultiByte(t *testing.T) {
+	input := `FROM test
+	SYSTEM 你好👋`
+
+	expect := []Command{
+		{Name: "model", Args: "test"},
+		{Name: "system", Args: "你好👋"},
+	}
+
+	encodings := []encoding.Encoding{
+		unicode.UTF8,
+		unicode.UTF16(unicode.LittleEndian, unicode.UseBOM),
+		unicode.UTF16(unicode.BigEndian, unicode.UseBOM),
+	}
+
+	for _, encoding := range encodings {
+		t.Run(fmt.Sprintf("%s", encoding), func(t *testing.T) {
+			s, err := encoding.NewEncoder().String(input)
+			require.NoError(t, err)
+
+			actual, err := ParseFile(strings.NewReader(s))
+			require.NoError(t, err)
+
+			assert.Equal(t, expect, actual.Commands)
+		})
+	}
 }

progress/progress.go

@@ -59,7 +59,7 @@ func (p *Progress) StopAndClear() bool {
 	stopped := p.stop()
 	if stopped {
 		// clear all progress lines
-		for i := 0; i < p.pos; i++ {
+		for i := range p.pos {
 			if i > 0 {
 				fmt.Fprint(p.w, "\033[A")
 			}
@@ -85,7 +85,7 @@ func (p *Progress) render() {
 	defer fmt.Fprint(p.w, "\033[?25h")
 
 	// clear already rendered progress lines
-	for i := 0; i < p.pos; i++ {
+	for i := range p.pos {
 		if i > 0 {
 			fmt.Fprint(p.w, "\033[A")
 		}

readline/buffer.go

@@ -52,7 +52,6 @@ func (b *Buffer) GetLineSpacing(line int) bool {
 	}
 
 	return hasSpace.(bool)
-
 }
 
 func (b *Buffer) MoveLeft() {
@@ -117,15 +116,12 @@ func (b *Buffer) MoveRight() {
 
 				if b.DisplayPos%b.LineWidth == 0 {
 					fmt.Printf(CursorDown + CursorBOL + cursorRightN(len(b.Prompt.prompt())))
-
 				} else if (b.DisplayPos-rLength)%b.LineWidth == b.LineWidth-1 && hasSpace {
 					fmt.Printf(CursorDown + CursorBOL + cursorRightN(len(b.Prompt.prompt())+rLength))
 					b.DisplayPos += 1
-
 				} else if b.LineHasSpace.Size() > 0 && b.DisplayPos%b.LineWidth == b.LineWidth-1 && hasSpace {
 					fmt.Printf(CursorDown + CursorBOL + cursorRightN(len(b.Prompt.prompt())))
 					b.DisplayPos += 1
-
 				} else {
 					fmt.Print(cursorRightN(rLength))
 				}
@@ -154,7 +150,7 @@ func (b *Buffer) MoveToStart() {
 	if b.Pos > 0 {
 		currLine := b.DisplayPos / b.LineWidth
 		if currLine > 0 {
-			for cnt := 0; cnt < currLine; cnt++ {
+			for range currLine {
 				fmt.Print(CursorUp)
 			}
 		}
@@ -169,7 +165,7 @@ func (b *Buffer) MoveToEnd() {
 		currLine := b.DisplayPos / b.LineWidth
 		totalLines := b.DisplaySize() / b.LineWidth
 		if currLine < totalLines {
-			for cnt := 0; cnt < totalLines-currLine; cnt++ {
+			for range totalLines - currLine {
 				fmt.Print(CursorDown)
 			}
 			remainder := b.DisplaySize() % b.LineWidth
@@ -185,7 +181,7 @@ func (b *Buffer) MoveToEnd() {
 
 func (b *Buffer) DisplaySize() int {
 	sum := 0
-	for i := 0; i < b.Buf.Size(); i++ {
+	for i := range b.Buf.Size() {
 		if e, ok := b.Buf.Get(i); ok {
 			if r, ok := e.(rune); ok {
 				sum += runewidth.RuneWidth(r)
@@ -197,7 +193,6 @@ func (b *Buffer) DisplaySize() int {
 }
 
 func (b *Buffer) Add(r rune) {
-
 	if b.Pos == b.Buf.Size() {
 		b.AddChar(r, false)
 	} else {
@@ -210,7 +205,6 @@ func (b *Buffer) AddChar(r rune, insert bool) {
 	b.DisplayPos += rLength
 
 	if b.Pos > 0 {
-
 		if b.DisplayPos%b.LineWidth == 0 {
 			fmt.Printf("%c", r)
 			fmt.Printf("\n%s", b.Prompt.AltPrompt)
@@ -235,7 +229,6 @@ func (b *Buffer) AddChar(r rune, insert bool) {
 			} else {
 				b.LineHasSpace.Add(true)
 			}
-
 		} else {
 			fmt.Printf("%c", r)
 		}
@@ -356,7 +349,6 @@ func (b *Buffer) drawRemaining() {
 
 func (b *Buffer) Remove() {
 	if b.Buf.Size() > 0 && b.Pos > 0 {
-
 		if e, ok := b.Buf.Get(b.Pos - 1); ok {
 			if r, ok := e.(rune); ok {
 				rLength := runewidth.RuneWidth(r)
@@ -382,7 +374,6 @@ func (b *Buffer) Remove() {
 					} else {
 						fmt.Print(" " + CursorLeft)
 					}
-
 				} else if (b.DisplayPos-rLength)%b.LineWidth == 0 && hasSpace {
 					fmt.Printf(CursorBOL + ClearToEOL)
 					fmt.Printf(CursorUp + CursorBOL + cursorRightN(b.Width))
@@ -391,10 +382,9 @@ func (b *Buffer) Remove() {
 						b.LineHasSpace.Remove(b.DisplayPos/b.LineWidth - 1)
 					}
 					b.DisplayPos -= 1
-
 				} else {
 					fmt.Print(cursorLeftN(rLength))
-					for i := 0; i < rLength; i++ {
+					for range rLength {
 						fmt.Print(" ")
 					}
 					fmt.Print(cursorLeftN(rLength))
@@ -451,7 +441,7 @@ func (b *Buffer) DeleteBefore() {
 func (b *Buffer) DeleteRemaining() {
 	if b.DisplaySize() > 0 && b.Pos < b.DisplaySize() {
 		charsToDel := b.Buf.Size() - b.Pos
-		for cnt := 0; cnt < charsToDel; cnt++ {
+		for range charsToDel {
 			b.Delete()
 		}
 	}
@@ -495,7 +485,7 @@ func (b *Buffer) ClearScreen() {
 		if currPos > 0 {
 			targetLine := currPos / b.LineWidth
 			if targetLine > 0 {
-				for cnt := 0; cnt < targetLine; cnt++ {
+				for range targetLine {
 					fmt.Print(CursorDown)
 				}
 			}
@@ -525,7 +515,7 @@ func (b *Buffer) Replace(r []rune) {
 
 	fmt.Printf(CursorBOL + ClearToEOL)
 
-	for i := 0; i < lineNums; i++ {
+	for range lineNums {
 		fmt.Print(CursorUp + CursorBOL + ClearToEOL)
 	}
 

readline/history.go

@@ -91,7 +91,7 @@ func (h *History) Add(l []rune) {
 func (h *History) Compact() {
 	s := h.Buf.Size()
 	if s > h.Limit {
-		for cnt := 0; cnt < s-h.Limit; cnt++ {
+		for range s - h.Limit {
 			h.Buf.Remove(0)
 		}
 	}
@@ -139,7 +139,7 @@ func (h *History) Save() error {
 	defer f.Close()
 
 	buf := bufio.NewWriter(f)
-	for cnt := 0; cnt < h.Size(); cnt++ {
+	for cnt := range h.Size() {
 		v, _ := h.Buf.Get(cnt)
 		line, _ := v.([]rune)
 		if _, err := buf.WriteString(string(line) + "\n"); err != nil {

readline/readline.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"io"
 	"os"
-	"syscall"
 )
 
 type Prompt struct {
@@ -63,7 +62,7 @@ func New(prompt Prompt) (*Instance, error) {
 
 func (i *Instance) Readline() (string, error) {
 	if !i.Terminal.rawmode {
-		fd := int(syscall.Stdin)
+		fd := os.Stdin.Fd()
 		termios, err := SetRawMode(fd)
 		if err != nil {
 			return "", err
@@ -80,8 +79,8 @@ func (i *Instance) Readline() (string, error) {
 	fmt.Print(prompt)
 
 	defer func() {
-		fd := int(syscall.Stdin)
-		// nolint: errcheck
+		fd := os.Stdin.Fd()
+		//nolint:errcheck
 		UnsetRawMode(fd, i.Terminal.termios)
 		i.Terminal.rawmode = false
 	}()
@@ -136,7 +135,7 @@ func (i *Instance) Readline() (string, error) {
 				buf.MoveRight()
 			case CharBracketedPaste:
 				var code string
-				for cnt := 0; cnt < 3; cnt++ {
+				for range 3 {
 					r, err = i.Terminal.Read()
 					if err != nil {
 						return "", io.EOF
@@ -198,7 +197,7 @@ func (i *Instance) Readline() (string, error) {
 			buf.Remove()
 		case CharTab:
 			// todo: convert back to real tabs
-			for cnt := 0; cnt < 8; cnt++ {
+			for range 8 {
 				buf.Add(' ')
 			}
 		case CharDelete:
@@ -216,7 +215,7 @@ func (i *Instance) Readline() (string, error) {
 		case CharCtrlW:
 			buf.DeleteWord()
 		case CharCtrlZ:
-			fd := int(syscall.Stdin)
+			fd := os.Stdin.Fd()
 			return handleCharCtrlZ(fd, i.Terminal.termios)
 		case CharEnter, CharCtrlJ:
 			output := buf.String()
@@ -248,7 +247,7 @@ func (i *Instance) HistoryDisable() {
 }
 
 func NewTerminal() (*Terminal, error) {
-	fd := int(syscall.Stdin)
+	fd := os.Stdin.Fd()
 	termios, err := SetRawMode(fd)
 	if err != nil {
 		return nil, err

readline/readline_unix.go

@@ -6,7 +6,7 @@ import (
 	"syscall"
 )
 
-func handleCharCtrlZ(fd int, termios any) (string, error) {
+func handleCharCtrlZ(fd uintptr, termios any) (string, error) {
 	t := termios.(*Termios)
 	if err := UnsetRawMode(fd, t); err != nil {
 		return "", err

readline/readline_windows.go

@@ -1,6 +1,6 @@
 package readline
 
-func handleCharCtrlZ(fd int, state any) (string, error) {
+func handleCharCtrlZ(fd uintptr, state any) (string, error) {
 	// not supported
 	return "", nil
 }

readline/term.go

@@ -8,7 +8,7 @@ import (
 
 type Termios syscall.Termios
 
-func SetRawMode(fd int) (*Termios, error) {
+func SetRawMode(fd uintptr) (*Termios, error) {
 	termios, err := getTermios(fd)
 	if err != nil {
 		return nil, err
@@ -25,13 +25,13 @@ func SetRawMode(fd int) (*Termios, error) {
 	return termios, setTermios(fd, &newTermios)
 }
 
-func UnsetRawMode(fd int, termios any) error {
+func UnsetRawMode(fd uintptr, termios any) error {
 	t := termios.(*Termios)
 	return setTermios(fd, t)
 }
 
 // IsTerminal returns true if the given file descriptor is a terminal.
-func IsTerminal(fd int) bool {
+func IsTerminal(fd uintptr) bool {
 	_, err := getTermios(fd)
 	return err == nil
 }

readline/term_bsd.go

@@ -7,17 +7,17 @@ import (
 	"unsafe"
 )
 
-func getTermios(fd int) (*Termios, error) {
+func getTermios(fd uintptr) (*Termios, error) {
 	termios := new(Termios)
-	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, uintptr(fd), syscall.TIOCGETA, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
+	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, fd, syscall.TIOCGETA, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
 	if err != 0 {
 		return nil, err
 	}
 	return termios, nil
 }
 
-func setTermios(fd int, termios *Termios) error {
-	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, uintptr(fd), syscall.TIOCSETA, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
+func setTermios(fd uintptr, termios *Termios) error {
+	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, fd, syscall.TIOCSETA, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
 	if err != 0 {
 		return err
 	}

readline/term_linux.go

@@ -10,17 +10,17 @@ import (
 const tcgets = 0x5401
 const tcsets = 0x5402
 
-func getTermios(fd int) (*Termios, error) {
+func getTermios(fd uintptr) (*Termios, error) {
 	termios := new(Termios)
-	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, uintptr(fd), tcgets, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
+	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, fd, tcgets, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
 	if err != 0 {
 		return nil, err
 	}
 	return termios, nil
 }
 
-func setTermios(fd int, termios *Termios) error {
-	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, uintptr(fd), tcsets, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
+func setTermios(fd uintptr, termios *Termios) error {
+	_, _, err := syscall.Syscall6(syscall.SYS_IOCTL, fd, tcsets, uintptr(unsafe.Pointer(termios)), 0, 0, 0)
 	if err != 0 {
 		return err
 	}

readline/term_windows.go

@@ -9,13 +9,13 @@ type State struct {
 }
 
 // IsTerminal checks if the given file descriptor is associated with a terminal
-func IsTerminal(fd int) bool {
+func IsTerminal(fd uintptr) bool {
 	var st uint32
 	err := windows.GetConsoleMode(windows.Handle(fd), &st)
 	return err == nil
 }
 
-func SetRawMode(fd int) (*State, error) {
+func SetRawMode(fd uintptr) (*State, error) {
 	var st uint32
 	if err := windows.GetConsoleMode(windows.Handle(fd), &st); err != nil {
 		return nil, err
@@ -32,7 +32,7 @@ func SetRawMode(fd int) (*State, error) {
 	return &State{st}, nil
 }
 
-func UnsetRawMode(fd int, state any) error {
+func UnsetRawMode(fd uintptr, state any) error {
 	s := state.(*State)
 	return windows.SetConsoleMode(windows.Handle(fd), s.mode)
 }

server/download.go

@@ -340,17 +340,17 @@ type downloadOpts struct {
 }
 
 // downloadBlob downloads a blob from the registry and stores it in the blobs directory
-func downloadBlob(ctx context.Context, opts downloadOpts) error {
+func downloadBlob(ctx context.Context, opts downloadOpts) (cacheHit bool, _ error) {
 	fp, err := GetBlobsPath(opts.digest)
 	if err != nil {
-		return err
+		return false, err
 	}
 
 	fi, err := os.Stat(fp)
 	switch {
 	case errors.Is(err, os.ErrNotExist):
 	case err != nil:
-		return err
+		return false, err
 	default:
 		opts.fn(api.ProgressResponse{
 			Status:    fmt.Sprintf("pulling %s", opts.digest[7:19]),
@@ -359,7 +359,7 @@ func downloadBlob(ctx context.Context, opts downloadOpts) error {
 			Completed: fi.Size(),
 		})
 
-		return nil
+		return true, nil
 	}
 
 	data, ok := blobDownloadManager.LoadOrStore(opts.digest, &blobDownload{Name: fp, Digest: opts.digest})
@@ -369,12 +369,12 @@ func downloadBlob(ctx context.Context, opts downloadOpts) error {
 		requestURL = requestURL.JoinPath("v2", opts.mp.GetNamespaceRepository(), "blobs", opts.digest)
 		if err := download.Prepare(ctx, requestURL, opts.regOpts); err != nil {
 			blobDownloadManager.Delete(opts.digest)
-			return err
+			return false, err
 		}
 
-		// nolint: contextcheck
+		//nolint:contextcheck
 		go download.Run(context.Background(), requestURL, opts.regOpts)
 	}
 
-	return download.Wait(ctx, opts.fn)
+	return false, download.Wait(ctx, opts.fn)
 }

server/images.go

@@ -18,17 +18,16 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"slices"
 	"strconv"
 	"strings"
 
-	"golang.org/x/exp/slices"
-
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/auth"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/llm"
 	"github.com/ollama/ollama/parser"
-	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/types/errtypes"
 	"github.com/ollama/ollama/types/model"
 	"github.com/ollama/ollama/version"
@@ -315,7 +314,7 @@ func realpath(rel, from string) string {
 	return abspath
 }
 
-func CreateModel(ctx context.Context, name, modelFileDir, quantization string, modelfile *parser.File, fn func(resp api.ProgressResponse)) (err error) {
+func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantization string, modelfile *parser.File, fn func(resp api.ProgressResponse)) (err error) {
 	config := ConfigV2{
 		OS:           "linux",
 		Architecture: "amd64",
@@ -333,7 +332,7 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 
 		switch c.Name {
 		case "model", "adapter":
-			var baseLayers []*layerWithGGML
+			var baseLayers []*layerGGML
 			if name := model.ParseName(c.Args); name.IsValid() {
 				baseLayers, err = parseFromModel(ctx, name, fn)
 				if err != nil {
@@ -440,19 +439,27 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 				layers = append(layers, baseLayer.Layer)
 			}
 		case "license", "template", "system":
+			if c.Name != "license" {
+				// replace
+				layers = slices.DeleteFunc(layers, func(layer *Layer) bool {
+					if layer.MediaType != mediatype {
+						return false
+					}
+
+					if err := layer.Remove(); err != nil {
+						return false
+					}
+
+					return true
+				})
+			}
+
 			blob := strings.NewReader(c.Args)
 			layer, err := NewLayer(blob, mediatype)
 			if err != nil {
 				return err
 			}
 
-			if c.Name != "license" {
-				// replace
-				layers = slices.DeleteFunc(layers, func(layer *Layer) bool {
-					return layer.MediaType == mediatype
-				})
-			}
-
 			layers = append(layers, layer)
 		case "message":
 			role, content, ok := strings.Cut(c.Args, ": ")
@@ -571,26 +578,15 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 		}
 	}
 
-	unref := make(map[string]struct{})
-	if manifest, _, err := GetManifest(ParseModelPath(name)); err == nil {
-		for _, layer := range manifest.Layers {
-			if !slices.Contains(digests, layer.Digest) {
-				unref[layer.Digest] = struct{}{}
-			}
-		}
-
-		if manifest.Config.Digest != layer.Digest {
-			unref[manifest.Config.Digest] = struct{}{}
-		}
-	}
+	old, _ := ParseNamedManifest(name)
 
 	fn(api.ProgressResponse{Status: "writing manifest"})
 	if err := WriteManifest(name, layer, layers); err != nil {
 		return err
 	}
 
-	if !envconfig.NoPrune {
-		if err := deleteUnusedLayers(nil, unref); err != nil {
+	if !envconfig.NoPrune && old != nil {
+		if err := old.RemoveLayers(); err != nil {
 			return err
 		}
 	}
@@ -662,7 +658,7 @@ func deleteUnusedLayers(skipModelPath *ModelPath, deleteMap map[string]struct{})
 		// save (i.e. delete from the deleteMap) any files used in other manifests
 		manifest, _, err := GetManifest(fmp)
 		if err != nil {
-			// nolint: nilerr
+			//nolint:nilerr
 			return nil
 		}
 
@@ -771,37 +767,6 @@ func PruneDirectory(path string) error {
 	return nil
 }
 
-func DeleteModel(name string) error {
-	mp := ParseModelPath(name)
-	manifest, _, err := GetManifest(mp)
-	if err != nil {
-		return err
-	}
-
-	deleteMap := make(map[string]struct{})
-	for _, layer := range manifest.Layers {
-		deleteMap[layer.Digest] = struct{}{}
-	}
-	deleteMap[manifest.Config.Digest] = struct{}{}
-
-	err = deleteUnusedLayers(&mp, deleteMap)
-	if err != nil {
-		return err
-	}
-
-	fp, err := mp.GetManifestPath()
-	if err != nil {
-		return err
-	}
-	err = os.Remove(fp)
-	if err != nil {
-		slog.Info(fmt.Sprintf("couldn't remove manifest file '%s': %v", fp, err))
-		return err
-	}
-
-	return nil
-}
-
 func PushModel(ctx context.Context, name string, regOpts *registryOptions, fn func(api.ProgressResponse)) error {
 	mp := ParseModelPath(name)
 	fn(api.ProgressResponse{Status: "retrieving manifest"})
@@ -888,23 +853,27 @@ func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn fu
 	layers = append(layers, manifest.Layers...)
 	layers = append(layers, manifest.Config)
 
+	skipVerify := make(map[string]bool)
 	for _, layer := range layers {
-		if err := downloadBlob(
-			ctx,
-			downloadOpts{
-				mp:      mp,
-				digest:  layer.Digest,
-				regOpts: regOpts,
-				fn:      fn,
-			}); err != nil {
+		cacheHit, err := downloadBlob(ctx, downloadOpts{
+			mp:      mp,
+			digest:  layer.Digest,
+			regOpts: regOpts,
+			fn:      fn,
+		})
+		if err != nil {
 			return err
 		}
+		skipVerify[layer.Digest] = cacheHit
 		delete(deleteMap, layer.Digest)
 	}
 	delete(deleteMap, manifest.Config.Digest)
 
 	fn(api.ProgressResponse{Status: "verifying sha256 digest"})
 	for _, layer := range layers {
+		if skipVerify[layer.Digest] {
+			continue
+		}
 		if err := verifyBlob(layer.Digest); err != nil {
 			if errors.Is(err, errDigestMismatch) {
 				// something went wrong, delete the blob
@@ -991,7 +960,6 @@ var errUnauthorized = fmt.Errorf("unauthorized: access denied")
 func getTokenSubject(token string) string {
 	parts := strings.Split(token, ".")
 	if len(parts) != 3 {
-		slog.Error("jwt token does not contain 3 parts")
 		return ""
 	}
 
@@ -1019,7 +987,7 @@ func getTokenSubject(token string) string {
 
 func makeRequestWithRetry(ctx context.Context, method string, requestURL *url.URL, headers http.Header, body io.ReadSeeker, regOpts *registryOptions) (*http.Response, error) {
 	anonymous := true // access will default to anonymous if no user is found associated with the public key
-	for i := 0; i < 2; i++ {
+	for range 2 {
 		resp, err := makeRequest(ctx, method, requestURL, headers, body, regOpts)
 		if err != nil {
 			if !errors.Is(err, context.Canceled) {

server/layer.go

@@ -88,3 +88,26 @@ func (l *Layer) Open() (io.ReadSeekCloser, error) {
 
 	return os.Open(blob)
 }
+
+func (l *Layer) Remove() error {
+	ms, err := Manifests()
+	if err != nil {
+		return err
+	}
+
+	for _, m := range ms {
+		for _, layer := range append(m.Layers, m.Config) {
+			if layer.Digest == l.Digest {
+				// something is using this layer
+				return nil
+			}
+		}
+	}
+
+	blob, err := GetBlobsPath(l.Digest)
+	if err != nil {
+		return err
+	}
+
+	return os.Remove(blob)
+}

server/manifest.go

@@ -1,11 +1,12 @@
 package server
 
 import (
-	"bytes"
 	"crypto/sha256"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"os"
 	"path/filepath"
 
@@ -14,7 +15,10 @@ import (
 
 type Manifest struct {
 	ManifestV2
-	Digest string `json:"-"`
+
+	filepath string
+	fi       os.FileInfo
+	digest   string
 }
 
 func (m *Manifest) Size() (size int64) {
@@ -25,9 +29,34 @@ func (m *Manifest) Size() (size int64) {
 	return
 }
 
-func ParseNamedManifest(name model.Name) (*Manifest, error) {
-	if !name.IsFullyQualified() {
-		return nil, model.Unqualified(name)
+func (m *Manifest) Remove() error {
+	if err := os.Remove(m.filepath); err != nil {
+		return err
+	}
+
+	manifests, err := GetManifestPath()
+	if err != nil {
+		return err
+	}
+
+	return PruneDirectory(manifests)
+}
+
+func (m *Manifest) RemoveLayers() error {
+	for _, layer := range append(m.Layers, m.Config) {
+		if err := layer.Remove(); errors.Is(err, os.ErrNotExist) {
+			slog.Debug("layer does not exist", "digest", layer.Digest)
+		} else if err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func ParseNamedManifest(n model.Name) (*Manifest, error) {
+	if !n.IsFullyQualified() {
+		return nil, model.Unqualified(n)
 	}
 
 	manifests, err := GetManifestPath()
@@ -35,45 +64,101 @@ func ParseNamedManifest(name model.Name) (*Manifest, error) {
 		return nil, err
 	}
 
-	var manifest ManifestV2
-	manifestfile, err := os.Open(filepath.Join(manifests, name.Filepath()))
+	p := filepath.Join(manifests, n.Filepath())
+
+	var m ManifestV2
+	f, err := os.Open(p)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	fi, err := f.Stat()
 	if err != nil {
 		return nil, err
 	}
 
 	sha256sum := sha256.New()
-	if err := json.NewDecoder(io.TeeReader(manifestfile, sha256sum)).Decode(&manifest); err != nil {
+	if err := json.NewDecoder(io.TeeReader(f, sha256sum)).Decode(&m); err != nil {
 		return nil, err
 	}
 
 	return &Manifest{
-		ManifestV2: manifest,
-		Digest:     fmt.Sprintf("%x", sha256sum.Sum(nil)),
+		ManifestV2: m,
+		filepath:   p,
+		fi:         fi,
+		digest:     fmt.Sprintf("%x", sha256sum.Sum(nil)),
 	}, nil
 }
 
-func WriteManifest(name string, config *Layer, layers []*Layer) error {
-	manifest := ManifestV2{
+func WriteManifest(name model.Name, config *Layer, layers []*Layer) error {
+	manifests, err := GetManifestPath()
+	if err != nil {
+		return err
+	}
+
+	p := filepath.Join(manifests, name.Filepath())
+	if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+		return err
+	}
+
+	f, err := os.Create(p)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+
+	m := ManifestV2{
 		SchemaVersion: 2,
 		MediaType:     "application/vnd.docker.distribution.manifest.v2+json",
 		Config:        config,
 		Layers:        layers,
 	}
 
-	var b bytes.Buffer
-	if err := json.NewEncoder(&b).Encode(manifest); err != nil {
-		return err
-	}
-
-	modelpath := ParseModelPath(name)
-	manifestPath, err := modelpath.GetManifestPath()
-	if err != nil {
-		return err
-	}
-
-	if err := os.MkdirAll(filepath.Dir(manifestPath), 0o755); err != nil {
-		return err
-	}
-
-	return os.WriteFile(manifestPath, b.Bytes(), 0o644)
+	return json.NewEncoder(f).Encode(m)
+}
+
+func Manifests() (map[model.Name]*Manifest, error) {
+	manifests, err := GetManifestPath()
+	if err != nil {
+		return nil, err
+	}
+
+	// TODO(mxyng): use something less brittle
+	matches, err := filepath.Glob(filepath.Join(manifests, "*", "*", "*", "*"))
+	if err != nil {
+		return nil, err
+	}
+
+	ms := make(map[model.Name]*Manifest)
+	for _, match := range matches {
+		fi, err := os.Stat(match)
+		if err != nil {
+			return nil, err
+		}
+
+		if !fi.IsDir() {
+			rel, err := filepath.Rel(manifests, match)
+			if err != nil {
+				slog.Warn("bad filepath", "path", match, "error", err)
+				continue
+			}
+
+			n := model.ParseNameFromFilepath(rel)
+			if !n.IsValid() {
+				slog.Warn("bad manifest name", "path", rel, "error", err)
+				continue
+			}
+
+			m, err := ParseNamedManifest(n)
+			if err != nil {
+				slog.Warn("bad manifest", "name", n, "error", err)
+				continue
+			}
+
+			ms[n] = m
+		}
+	}
+
+	return ms, nil
 }

server/manifest_test.go

@@ -0,0 +1,152 @@
+package server
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"slices"
+	"testing"
+
+	"github.com/ollama/ollama/envconfig"
+	"github.com/ollama/ollama/types/model"
+)
+
+func createManifest(t *testing.T, path, name string) {
+	t.Helper()
+
+	p := filepath.Join(path, "manifests", name)
+	if err := os.MkdirAll(filepath.Dir(p), 0755); err != nil {
+		t.Fatal(err)
+	}
+
+	f, err := os.Create(p)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+
+	if err := json.NewEncoder(f).Encode(ManifestV2{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestManifests(t *testing.T) {
+	cases := map[string]struct {
+		ps               []string
+		wantValidCount   int
+		wantInvalidCount int
+	}{
+		"empty": {},
+		"single": {
+			ps: []string{
+				filepath.Join("host", "namespace", "model", "tag"),
+			},
+			wantValidCount: 1,
+		},
+		"multiple": {
+			ps: []string{
+				filepath.Join("registry.ollama.ai", "library", "llama3", "latest"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q4_0"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q4_1"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q8_0"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q5_0"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q5_1"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q2_K"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q3_K_S"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q3_K_M"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q3_K_L"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q4_K_S"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q4_K_M"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q5_K_S"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q5_K_M"),
+				filepath.Join("registry.ollama.ai", "library", "llama3", "q6_K"),
+			},
+			wantValidCount: 15,
+		},
+		"hidden": {
+			ps: []string{
+				filepath.Join("host", "namespace", "model", "tag"),
+				filepath.Join("host", "namespace", "model", ".hidden"),
+			},
+			wantValidCount:   1,
+			wantInvalidCount: 1,
+		},
+		"subdir": {
+			ps: []string{
+				filepath.Join("host", "namespace", "model", "tag", "one"),
+				filepath.Join("host", "namespace", "model", "tag", "another", "one"),
+			},
+			wantInvalidCount: 2,
+		},
+		"upper tag": {
+			ps: []string{
+				filepath.Join("host", "namespace", "model", "TAG"),
+			},
+			wantValidCount: 1,
+		},
+		"upper model": {
+			ps: []string{
+				filepath.Join("host", "namespace", "MODEL", "tag"),
+			},
+			wantValidCount: 1,
+		},
+		"upper namespace": {
+			ps: []string{
+				filepath.Join("host", "NAMESPACE", "model", "tag"),
+			},
+			wantValidCount: 1,
+		},
+		"upper host": {
+			ps: []string{
+				filepath.Join("HOST", "namespace", "model", "tag"),
+			},
+			wantValidCount: 1,
+		},
+	}
+
+	for n, wants := range cases {
+		t.Run(n, func(t *testing.T) {
+			d := t.TempDir()
+			t.Setenv("OLLAMA_MODELS", d)
+			envconfig.LoadConfig()
+
+			for _, p := range wants.ps {
+				createManifest(t, d, p)
+			}
+
+			ms, err := Manifests()
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			var ns []model.Name
+			for k := range ms {
+				ns = append(ns, k)
+			}
+
+			var gotValidCount, gotInvalidCount int
+			for _, p := range wants.ps {
+				n := model.ParseNameFromFilepath(p)
+				if n.IsValid() {
+					gotValidCount++
+				} else {
+					gotInvalidCount++
+				}
+
+				if !n.IsValid() && slices.Contains(ns, n) {
+					t.Errorf("unexpected invalid name: %s", p)
+				} else if n.IsValid() && !slices.Contains(ns, n) {
+					t.Errorf("missing valid name: %s", p)
+				}
+			}
+
+			if gotValidCount != wants.wantValidCount {
+				t.Errorf("got valid count %d, want %d", gotValidCount, wants.wantValidCount)
+			}
+
+			if gotInvalidCount != wants.wantInvalidCount {
+				t.Errorf("got invalid count %d, want %d", gotInvalidCount, wants.wantInvalidCount)
+			}
+		})
+	}
+}

server/model.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
@@ -14,27 +15,26 @@ import (
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/convert"
 	"github.com/ollama/ollama/llm"
+	"github.com/ollama/ollama/templates"
 	"github.com/ollama/ollama/types/model"
 )
 
 var intermediateBlobs map[string]string = make(map[string]string)
 
-type layerWithGGML struct {
+type layerGGML 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)
+func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
+	m, err := ParseNamedManifest(name)
 	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)
+		m, err = ParseNamedManifest(name)
 		if err != nil {
 			return nil, err
 		}
@@ -42,8 +42,8 @@ func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressRe
 		return nil, err
 	}
 
-	for _, layer := range manifest.Layers {
-		layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, modelpath.GetShortTagname())
+	for _, layer := range m.Layers {
+		layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, name.DisplayShortest())
 		if err != nil {
 			return nil, err
 		}
@@ -68,17 +68,16 @@ func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressRe
 				return nil, err
 			}
 
-			layers = append(layers, &layerWithGGML{layer, ggml})
+			layers = append(layers, &layerGGML{layer, ggml})
 		default:
-			layers = append(layers, &layerWithGGML{layer, nil})
+			layers = append(layers, &layerGGML{layer, nil})
 		}
-
 	}
 
 	return layers, nil
 }
 
-func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
+func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
 	stat, err := file.Stat()
 	if err != nil {
 		return nil, err
@@ -182,13 +181,13 @@ func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(a
 		return nil, err
 	}
 
-	layers = append(layers, &layerWithGGML{layer, ggml})
+	layers = append(layers, &layerGGML{layer, ggml})
 
 	intermediateBlobs[digest] = layer.Digest
-	return layers, nil
+	return detectChatTemplate(layers)
 }
 
-func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
+func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
 	sr := io.NewSectionReader(file, 0, 512)
 	contentType, err := detectContentType(sr)
 	if err != nil {
@@ -230,10 +229,30 @@ func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(ap
 			return nil, err
 		}
 
-		layers = append(layers, &layerWithGGML{layer, ggml})
+		layers = append(layers, &layerGGML{layer, ggml})
 		offset = n
 	}
 
+	return detectChatTemplate(layers)
+}
+
+func detectChatTemplate(layers []*layerGGML) ([]*layerGGML, error) {
+	for _, layer := range layers {
+		if s := layer.GGML.KV().ChatTemplate(); s != "" {
+			if t, err := templates.NamedTemplate(s); err != nil {
+				slog.Debug("template detection", "error", err)
+			} else {
+				tmpl, err := NewLayer(t.Reader(), "application/vnd.ollama.image.template")
+				if err != nil {
+					return nil, err
+				}
+
+				tmpl.status = fmt.Sprintf("using autodetected template %s", t.Name)
+				layers = append(layers, &layerGGML{tmpl, nil})
+			}
+		}
+	}
+
 	return layers, nil
 }
 

server/modelpath.go

@@ -8,6 +8,8 @@ import (
 	"path/filepath"
 	"regexp"
 	"strings"
+
+	"github.com/ollama/ollama/envconfig"
 )
 
 type ModelPath struct {
@@ -104,14 +106,7 @@ func (mp ModelPath) GetShortTagname() string {
 // modelsDir returns the value of the OLLAMA_MODELS environment variable or the user's home directory if OLLAMA_MODELS is not set.
 // The models directory is where Ollama stores its model files and manifests.
 func modelsDir() (string, error) {
-	if models, exists := os.LookupEnv("OLLAMA_MODELS"); exists {
-		return models, nil
-	}
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return "", err
-	}
-	return filepath.Join(home, ".ollama", "models"), nil
+	return envconfig.ModelsDir, nil
 }
 
 // GetManifestPath returns the path to the manifest file for the given model path, it is up to the caller to create the directory if it does not exist.

server/modelpath_test.go

@@ -6,12 +6,15 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/ollama/ollama/envconfig"
 )
 
 func TestGetBlobsPath(t *testing.T) {
 	// GetBlobsPath expects an actual directory to exist
 	dir, err := os.MkdirTemp("", "ollama-test")
-	assert.Nil(t, err)
+	require.NoError(t, err)
 	defer os.RemoveAll(dir)
 
 	tests := []struct {
@@ -60,10 +63,11 @@ func TestGetBlobsPath(t *testing.T) {
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
 			t.Setenv("OLLAMA_MODELS", dir)
+			envconfig.LoadConfig()
 
 			got, err := GetBlobsPath(tc.digest)
 
-			assert.ErrorIs(t, tc.err, err, tc.name)
+			require.ErrorIs(t, tc.err, err, tc.name)
 			assert.Equal(t, tc.expected, got, tc.name)
 		})
 	}

server/routes.go

@@ -16,6 +16,7 @@ import (
 	"os"
 	"os/signal"
 	"path/filepath"
+	"slices"
 	"strconv"
 	"strings"
 	"syscall"
@@ -23,7 +24,6 @@ import (
 
 	"github.com/gin-contrib/cors"
 	"github.com/gin-gonic/gin"
-	"golang.org/x/exp/slices"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/envconfig"
@@ -77,7 +77,6 @@ func isSupportedImageType(image []byte) bool {
 }
 
 func (s *Server) GenerateHandler(c *gin.Context) {
-
 	checkpointStart := time.Now()
 	var req api.GenerateRequest
 	err := c.ShouldBindJSON(&req)
@@ -421,13 +420,14 @@ func (s *Server) PullModelHandler(c *gin.Context) {
 		return
 	}
 
-	var model string
-	if req.Model != "" {
-		model = req.Model
-	} else if req.Name != "" {
-		model = req.Name
-	} else {
-		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
+	name := model.ParseName(cmp.Or(req.Model, req.Name))
+	if !name.IsValid() {
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "invalid model name"})
+		return
+	}
+
+	if err := checkNameExists(name); err != nil {
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
 		return
 	}
 
@@ -445,7 +445,7 @@ func (s *Server) PullModelHandler(c *gin.Context) {
 		ctx, cancel := context.WithCancel(c.Request.Context())
 		defer cancel()
 
-		if err := PullModel(ctx, model, regOpts, fn); err != nil {
+		if err := PullModel(ctx, name.DisplayShortest(), regOpts, fn); err != nil {
 			ch <- gin.H{"error": err.Error()}
 		}
 	}()
@@ -507,9 +507,24 @@ func (s *Server) PushModelHandler(c *gin.Context) {
 	streamResponse(c, ch)
 }
 
+func checkNameExists(name model.Name) error {
+	names, err := Manifests()
+	if err != nil {
+		return err
+	}
+
+	for n := range names {
+		if strings.EqualFold(n.Filepath(), name.Filepath()) && n != name {
+			return fmt.Errorf("a model with that name already exists")
+		}
+	}
+
+	return nil
+}
+
 func (s *Server) CreateModelHandler(c *gin.Context) {
-	var req api.CreateRequest
-	if err := c.ShouldBindJSON(&req); errors.Is(err, io.EOF) {
+	var r api.CreateRequest
+	if err := c.ShouldBindJSON(&r); errors.Is(err, io.EOF) {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
 		return
 	} else if err != nil {
@@ -517,30 +532,35 @@ func (s *Server) CreateModelHandler(c *gin.Context) {
 		return
 	}
 
-	name := model.ParseName(cmp.Or(req.Model, req.Name))
+	name := model.ParseName(cmp.Or(r.Model, r.Name))
 	if !name.IsValid() {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": errtypes.InvalidModelNameErrMsg})
 		return
 	}
 
-	if req.Path == "" && req.Modelfile == "" {
+	if err := checkNameExists(name); err != nil {
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+		return
+	}
+
+	if r.Path == "" && r.Modelfile == "" {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "path or modelfile are required"})
 		return
 	}
 
-	var r io.Reader = strings.NewReader(req.Modelfile)
-	if req.Path != "" && req.Modelfile == "" {
-		f, err := os.Open(req.Path)
+	var sr io.Reader = strings.NewReader(r.Modelfile)
+	if r.Path != "" && r.Modelfile == "" {
+		f, err := os.Open(r.Path)
 		if err != nil {
 			c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("error reading modelfile: %s", err)})
 			return
 		}
 		defer f.Close()
 
-		r = f
+		sr = f
 	}
 
-	modelfile, err := parser.ParseFile(r)
+	f, err := parser.ParseFile(sr)
 	if err != nil {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
 		return
@@ -556,17 +576,13 @@ func (s *Server) CreateModelHandler(c *gin.Context) {
 		ctx, cancel := context.WithCancel(c.Request.Context())
 		defer cancel()
 
-		quantization := req.Quantization
-		if req.Quantize != "" {
-			quantization = req.Quantize
-		}
-
-		if err := CreateModel(ctx, name.String(), filepath.Dir(req.Path), strings.ToUpper(quantization), modelfile, fn); err != nil {
+		quantization := cmp.Or(r.Quantize, r.Quantization)
+		if err := CreateModel(ctx, name, filepath.Dir(r.Path), strings.ToUpper(quantization), f, fn); err != nil {
 			ch <- gin.H{"error": err.Error()}
 		}
 	}()
 
-	if req.Stream != nil && !*req.Stream {
+	if r.Stream != nil && !*r.Stream {
 		waitForStream(c, ch)
 		return
 	}
@@ -575,48 +591,36 @@ func (s *Server) CreateModelHandler(c *gin.Context) {
 }
 
 func (s *Server) DeleteModelHandler(c *gin.Context) {
-	var req api.DeleteRequest
-	err := c.ShouldBindJSON(&req)
-	switch {
-	case errors.Is(err, io.EOF):
+	var r api.DeleteRequest
+	if err := c.ShouldBindJSON(&r); errors.Is(err, io.EOF) {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
 		return
-	case err != nil:
+	} else if err != nil {
 		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
 		return
 	}
 
-	var model string
-	if req.Model != "" {
-		model = req.Model
-	} else if req.Name != "" {
-		model = req.Name
-	} else {
-		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
+	n := model.ParseName(cmp.Or(r.Model, r.Name))
+	if !n.IsValid() {
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("name %q is invalid", cmp.Or(r.Model, r.Name))})
 		return
 	}
 
-	if err := DeleteModel(model); err != nil {
-		if os.IsNotExist(err) {
-			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", model)})
-		} else {
-			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
-		}
-		return
-	}
-
-	manifestsPath, err := GetManifestPath()
+	m, err := ParseNamedManifest(n)
 	if err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 		return
 	}
 
-	if err := PruneDirectory(manifestsPath); err != nil {
+	if err := m.Remove(); err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 		return
 	}
 
-	c.JSON(http.StatusOK, nil)
+	if err := m.RemoveLayers(); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
 }
 
 func (s *Server) ShowModelHandler(c *gin.Context) {
@@ -720,75 +724,45 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
 }
 
 func (s *Server) ListModelsHandler(c *gin.Context) {
-	manifests, err := GetManifestPath()
+	ms, err := Manifests()
 	if err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 		return
 	}
 
-	models := []api.ModelResponse{}
-	if err := filepath.Walk(manifests, func(path string, info os.FileInfo, _ error) error {
-		if !info.IsDir() {
-			rel, err := filepath.Rel(manifests, path)
-			if err != nil {
-				return err
-			}
+	models := []api.ListModelResponse{}
+	for n, m := range ms {
+		f, err := m.Config.Open()
+		if err != nil {
+			slog.Warn("bad manifest filepath", "name", n, "error", err)
+			continue
+		}
+		defer f.Close()
 
-			if hidden, err := filepath.Match(".*", filepath.Base(rel)); err != nil {
-				return err
-			} else if hidden {
-				return nil
-			}
-
-			n := model.ParseNameFromFilepath(rel)
-			if !n.IsValid() {
-				slog.Warn("bad manifest filepath", "path", rel)
-				return nil
-			}
-
-			m, err := ParseNamedManifest(n)
-			if err != nil {
-				slog.Warn("bad manifest", "name", n, "error", err)
-				return nil
-			}
-
-			f, err := m.Config.Open()
-			if err != nil {
-				slog.Warn("bad manifest config filepath", "name", n, "error", err)
-				return nil
-			}
-			defer f.Close()
-
-			var c ConfigV2
-			if err := json.NewDecoder(f).Decode(&c); err != nil {
-				slog.Warn("bad manifest config", "name", n, "error", err)
-				return nil
-			}
-
-			// tag should never be masked
-			models = append(models, api.ModelResponse{
-				Model:      n.DisplayShortest(),
-				Name:       n.DisplayShortest(),
-				Size:       m.Size(),
-				Digest:     m.Digest,
-				ModifiedAt: info.ModTime(),
-				Details: api.ModelDetails{
-					Format:            c.ModelFormat,
-					Family:            c.ModelFamily,
-					Families:          c.ModelFamilies,
-					ParameterSize:     c.ModelType,
-					QuantizationLevel: c.FileType,
-				},
-			})
+		var cf ConfigV2
+		if err := json.NewDecoder(f).Decode(&cf); err != nil {
+			slog.Warn("bad manifest config", "name", n, "error", err)
+			continue
 		}
 
-		return nil
-	}); err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
-		return
+		// tag should never be masked
+		models = append(models, api.ListModelResponse{
+			Model:      n.DisplayShortest(),
+			Name:       n.DisplayShortest(),
+			Size:       m.Size(),
+			Digest:     m.digest,
+			ModifiedAt: m.fi.ModTime(),
+			Details: api.ModelDetails{
+				Format:            cf.ModelFormat,
+				Family:            cf.ModelFamily,
+				Families:          cf.ModelFamilies,
+				ParameterSize:     cf.ModelType,
+				QuantizationLevel: cf.FileType,
+			},
+		})
 	}
 
-	slices.SortStableFunc(models, func(i, j api.ModelResponse) int {
+	slices.SortStableFunc(models, func(i, j api.ListModelResponse) int {
 		// most recently modified first
 		return cmp.Compare(j.ModifiedAt.Unix(), i.ModifiedAt.Unix())
 	})
@@ -818,6 +792,11 @@ func (s *Server) CopyModelHandler(c *gin.Context) {
 		return
 	}
 
+	if err := checkNameExists(dst); err != nil {
+		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+		return
+	}
+
 	if err := CopyModel(src, dst); errors.Is(err, os.ErrNotExist) {
 		c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model %q not found", r.Source)})
 	} else if err != nil {
@@ -963,7 +942,7 @@ func allowedHostsMiddleware(addr net.Addr) gin.HandlerFunc {
 		}
 
 		if allowedHost(host) {
-			if c.Request.Method == "OPTIONS" {
+			if c.Request.Method == http.MethodOptions {
 				c.AbortWithStatus(http.StatusNoContent)
 				return
 			}
@@ -981,6 +960,10 @@ func (s *Server) GenerateRoutes() http.Handler {
 	config.AllowWildcard = true
 	config.AllowBrowserExtensions = true
 	config.AllowHeaders = []string{"Authorization", "Content-Type", "User-Agent", "Accept", "X-Requested-With"}
+	openAIProperties := []string{"lang", "package-version", "os", "arch", "runtime", "runtime-version", "async"}
+	for _, prop := range openAIProperties {
+		config.AllowHeaders = append(config.AllowHeaders, "x-stainless-"+prop)
+	}
 	config.AllowOrigins = envconfig.AllowOrigins
 
 	r := gin.Default()
@@ -1004,6 +987,7 @@ func (s *Server) GenerateRoutes() http.Handler {
 
 	// Compatibility endpoints
 	r.POST("/v1/chat/completions", openai.Middleware(), s.ChatHandler)
+	r.DELETE("/v1/models/:model", openai.DeleteMiddleware(), s.DeleteModelHandler)
 
 	for _, method := range []string{http.MethodGet, http.MethodHead} {
 		r.Handle(method, "/", func(c *gin.Context) {
@@ -1160,7 +1144,7 @@ func streamResponse(c *gin.Context, ch chan any) {
 }
 
 func (s *Server) ProcessHandler(c *gin.Context) {
-	models := []api.ModelResponse{}
+	models := []api.ProcessModelResponse{}
 
 	for _, v := range s.sched.loaded {
 		model := v.model
@@ -1172,7 +1156,7 @@ func (s *Server) ProcessHandler(c *gin.Context) {
 			QuantizationLevel: model.Config.FileType,
 		}
 
-		mr := api.ModelResponse{
+		mr := api.ProcessModelResponse{
 			Model:     model.ShortName,
 			Name:      model.ShortName,
 			Size:      int64(v.estimatedTotal),
@@ -1192,7 +1176,7 @@ func (s *Server) ProcessHandler(c *gin.Context) {
 		models = append(models, mr)
 	}
 
-	c.JSON(http.StatusOK, api.ListResponse{Models: models})
+	c.JSON(http.StatusOK, api.ProcessResponse{Models: models})
 }
 
 // ChatPrompt builds up a prompt from a series of messages for the currently `loaded` model
@@ -1327,7 +1311,6 @@ func (s *Server) ChatHandler(c *gin.Context) {
 		defer close(ch)
 
 		fn := func(r llm.CompletionResponse) {
-
 			resp := api.ChatResponse{
 				Model:      req.Model,
 				CreatedAt:  time.Now().UTC(),

server/routes_create_test.go

@@ -0,0 +1,570 @@
+package server
+
+import (
+	"bytes"
+	"encoding/binary"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"slices"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
+	"github.com/ollama/ollama/llm"
+)
+
+var stream bool = false
+
+func createBinFile(t *testing.T, kv map[string]any, ti []llm.Tensor) string {
+	t.Helper()
+
+	f, err := os.CreateTemp(t.TempDir(), "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+
+	if err := llm.NewGGUFV3(binary.LittleEndian).Encode(f, kv, ti); err != nil {
+		t.Fatal(err)
+	}
+
+	return f.Name()
+}
+
+type responseRecorder struct {
+	*httptest.ResponseRecorder
+	http.CloseNotifier
+}
+
+func NewRecorder() *responseRecorder {
+	return &responseRecorder{
+		ResponseRecorder: httptest.NewRecorder(),
+	}
+}
+
+func (t *responseRecorder) CloseNotify() <-chan bool {
+	return make(chan bool)
+}
+
+func createRequest(t *testing.T, fn func(*gin.Context), body any) *httptest.ResponseRecorder {
+	t.Helper()
+
+	w := NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+
+	var b bytes.Buffer
+	if err := json.NewEncoder(&b).Encode(body); err != nil {
+		t.Fatal(err)
+	}
+
+	c.Request = &http.Request{
+		Body: io.NopCloser(&b),
+	}
+
+	fn(c)
+	return w.ResponseRecorder
+}
+
+func checkFileExists(t *testing.T, p string, expect []string) {
+	t.Helper()
+
+	actual, err := filepath.Glob(p)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if !slices.Equal(actual, expect) {
+		t.Fatalf("expected slices to be equal %v", actual)
+	}
+}
+
+func TestCreateFromBin(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+
+	var s Server
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-ca239d7bd8ea90e4a5d2e6bf88f8d74a47b14336e73eb4e18bed4dd325018116"),
+	})
+}
+
+func TestCreateFromModel(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test2",
+		Modelfile: "FROM test",
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-ca239d7bd8ea90e4a5d2e6bf88f8d74a47b14336e73eb4e18bed4dd325018116"),
+	})
+}
+
+func TestCreateRemovesLayers(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nTEMPLATE {{ .Prompt }}", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-b507b9c2f6ca642bffcd06665ea7c91f235fd32daeefdf875a0f938db05fb315"),
+		filepath.Join(p, "blobs", "sha256-bc80b03733773e0728011b2f4adf34c458b400e1aad48cb28d61170f3a2ad2d6"),
+	})
+
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nTEMPLATE {{ .System }} {{ .Prompt }}", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-8f2c2167d789c6b2302dff965160fa5029f6a24096d262c1cbb469f21a045382"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-fe7ac77b725cda2ccad03f88a880ecdfd7a33192d6cae08fce2c0ee1455991ed"),
+	})
+}
+
+func TestCreateUnsetsSystem(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nSYSTEM Say hi!", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-8585df945d1069bc78b79bd10bb73ba07fbc29b0f5479a31a601c0d12731416e"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-f29e82a8284dbdf5910b1555580ff60b04238b8da9d5e51159ada67a4d0d5851"),
+	})
+
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nSYSTEM \"\"", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-67d4b8d106af2a5b100a46e9bdc038c71eef2a35c9abac784092654212f97cf5"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"),
+	})
+
+	bts, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if string(bts) != "" {
+		t.Fatalf("expected empty string, actual %s", string(bts))
+	}
+}
+
+func TestCreateMergeParameters(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nPARAMETER temperature 1\nPARAMETER top_k 10\nPARAMETER stop USER:\nPARAMETER stop ASSISTANT:", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-1d0ad71299d48c2fb7ae2b98e683643e771f8a5b72be34942af90d97a91c1e37"),
+		filepath.Join(p, "blobs", "sha256-4a384beaf47a9cbe452dfa5ab70eea691790f3b35a832d12933a1996685bf2b6"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+	})
+
+	// in order to merge parameters, the second model must be created FROM the first
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test2",
+		Modelfile: "FROM test\nPARAMETER temperature 0.6\nPARAMETER top_p 0.7",
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-1d0ad71299d48c2fb7ae2b98e683643e771f8a5b72be34942af90d97a91c1e37"),
+		filepath.Join(p, "blobs", "sha256-4a384beaf47a9cbe452dfa5ab70eea691790f3b35a832d12933a1996685bf2b6"),
+		filepath.Join(p, "blobs", "sha256-4cd9d4ba6b734d9b4cbd1e5caa60374c00722e993fce5e1e2d15a33698f71187"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-e29a7b3c47287a2489c895d21fe413c20f859a85d20e749492f52a838e36e1ba"),
+	})
+
+	actual, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-e29a7b3c47287a2489c895d21fe413c20f859a85d20e749492f52a838e36e1ba"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	expect, err := json.Marshal(map[string]any{"temperature": 0.6, "top_k": 10, "top_p": 0.7, "stop": []string{"USER:", "ASSISTANT:"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if !bytes.Equal(bytes.TrimSpace(expect), bytes.TrimSpace(actual)) {
+		t.Errorf("expected %s, actual %s", string(expect), string(actual))
+	}
+
+	// slices are replaced
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test2",
+		Modelfile: "FROM test\nPARAMETER temperature 0.6\nPARAMETER top_p 0.7\nPARAMETER stop <|endoftext|>",
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-12f58bb75cb3042d69a7e013ab87fb3c3c7088f50ddc62f0c77bd332f0d44d35"),
+		filepath.Join(p, "blobs", "sha256-1d0ad71299d48c2fb7ae2b98e683643e771f8a5b72be34942af90d97a91c1e37"),
+		filepath.Join(p, "blobs", "sha256-257aa726584f24970a4f240765e75a7169bfbe7f4966c1f04513d6b6c860583a"),
+		filepath.Join(p, "blobs", "sha256-4a384beaf47a9cbe452dfa5ab70eea691790f3b35a832d12933a1996685bf2b6"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+	})
+
+	actual, err = os.ReadFile(filepath.Join(p, "blobs", "sha256-12f58bb75cb3042d69a7e013ab87fb3c3c7088f50ddc62f0c77bd332f0d44d35"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	expect, err = json.Marshal(map[string]any{"temperature": 0.6, "top_k": 10, "top_p": 0.7, "stop": []string{"<|endoftext|>"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if !bytes.Equal(bytes.TrimSpace(expect), bytes.TrimSpace(actual)) {
+		t.Errorf("expected %s, actual %s", string(expect), string(actual))
+	}
+}
+
+func TestCreateReplacesMessages(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nMESSAGE assistant \"What is my purpose?\"\nMESSAGE user \"You run tests.\"\nMESSAGE assistant \"Oh, my god.\"", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-298baeaf6928a60cf666d88d64a1ba606feb43a2865687c39e40652e407bffc4"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-e0e27d47045063ccb167ae852c51d49a98eab33fabaee4633fdddf97213e40b5"),
+	})
+
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test2",
+		Modelfile: "FROM test\nMESSAGE assistant \"You're a test, Harry.\"\nMESSAGE user \"I-I'm a what?\"\nMESSAGE assistant \"A test. And a thumping good one at that, I'd wager.\"",
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-298baeaf6928a60cf666d88d64a1ba606feb43a2865687c39e40652e407bffc4"),
+		filepath.Join(p, "blobs", "sha256-4f48b25fe9969564c82f58eb1cedbdff6484cc0baf474bc6c2a9b37c8da3362a"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-a60ecc9da299ec7ede453f99236e5577fd125e143689b646d9f0ddc9971bf4db"),
+		filepath.Join(p, "blobs", "sha256-e0e27d47045063ccb167ae852c51d49a98eab33fabaee4633fdddf97213e40b5"),
+	})
+
+	type message struct {
+		Role    string `json:"role"`
+		Content string `json:"content"`
+	}
+
+	f, err := os.Open(filepath.Join(p, "blobs", "sha256-a60ecc9da299ec7ede453f99236e5577fd125e143689b646d9f0ddc9971bf4db"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+
+	var actual []message
+	if err := json.NewDecoder(f).Decode(&actual); err != nil {
+		t.Fatal(err)
+	}
+
+	expect := []message{
+		{Role: "assistant", Content: "You're a test, Harry."},
+		{Role: "user", Content: "I-I'm a what?"},
+		{Role: "assistant", Content: "A test. And a thumping good one at that, I'd wager."},
+	}
+
+	if !slices.Equal(actual, expect) {
+		t.Errorf("expected %s, actual %s", expect, actual)
+	}
+}
+
+func TestCreateTemplateSystem(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nTEMPLATE {{ .Prompt }}\nSYSTEM Say hello!\nTEMPLATE {{ .System }} {{ .Prompt }}\nSYSTEM Say bye!", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-2b5e330885117c82f3fd75169ea323e141070a2947c11ddb9f79ee0b01c589c1"),
+		filepath.Join(p, "blobs", "sha256-4c5f51faac758fecaff8db42f0b7382891a4d0c0bb885f7b86be88c814a7cc86"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-fe7ac77b725cda2ccad03f88a880ecdfd7a33192d6cae08fce2c0ee1455991ed"),
+	})
+
+	template, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-fe7ac77b725cda2ccad03f88a880ecdfd7a33192d6cae08fce2c0ee1455991ed"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if string(template) != "{{ .System }} {{ .Prompt }}" {
+		t.Errorf("expected \"{{ .System }} {{ .Prompt }}\", actual %s", template)
+	}
+
+	system, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-4c5f51faac758fecaff8db42f0b7382891a4d0c0bb885f7b86be88c814a7cc86"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if string(system) != "Say bye!" {
+		t.Errorf("expected \"Say bye!\", actual %s", system)
+	}
+}
+
+func TestCreateLicenses(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s\nLICENSE MIT\nLICENSE Apache-2.0", createBinFile(t, nil, nil)),
+		Stream:    &stream,
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-2af71558e438db0b73a20beab92dc278a94e1bbe974c00c1a33e3ab62d53a608"),
+		filepath.Join(p, "blobs", "sha256-79a39c37536ddee29cbadd5d5e2dcba8ed7f03e431f626ff38432c1c866bb7e2"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-e5dcffe836b6ec8a58e492419b550e65fb8cbdc308503979e5dacb33ac7ea3b7"),
+	})
+
+	mit, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-e5dcffe836b6ec8a58e492419b550e65fb8cbdc308503979e5dacb33ac7ea3b7"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if string(mit) != "MIT" {
+		t.Errorf("expected MIT, actual %s", mit)
+	}
+
+	apache, err := os.ReadFile(filepath.Join(p, "blobs", "sha256-2af71558e438db0b73a20beab92dc278a94e1bbe974c00c1a33e3ab62d53a608"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if string(apache) != "Apache-2.0" {
+		t.Errorf("expected Apache-2.0, actual %s", apache)
+	}
+}
+
+func TestCreateDetectTemplate(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+	var s Server
+
+	t.Run("matched", func(t *testing.T) {
+		w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+			Name: "test",
+			Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, llm.KV{
+				"tokenizer.chat_template": "{{ bos_token }}{% for message in messages %}{{'<|' + message['role'] + '|>' + '\n' + message['content'] + '<|end|>\n' }}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>\n' }}{% else %}{{ eos_token }}{% endif %}",
+			}, nil)),
+			Stream: &stream,
+		})
+
+		if w.Code != http.StatusOK {
+			t.Fatalf("expected status code 200, actual %d", w.Code)
+		}
+
+		checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+			filepath.Join(p, "blobs", "sha256-2f8e594e6f34b1b4d36a246628eeb3365ce442303d656f1fcc69e821722acea0"),
+			filepath.Join(p, "blobs", "sha256-542b217f179c7825eeb5bca3c77d2b75ed05bafbd3451d9188891a60a85337c6"),
+			filepath.Join(p, "blobs", "sha256-553c4a3f747b3d22a4946875f1cc8ed011c2930d83f864a0c7265f9ec0a20413"),
+		})
+	})
+
+	t.Run("unmatched", func(t *testing.T) {
+		w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+			Name:      "test",
+			Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+			Stream:    &stream,
+		})
+
+		if w.Code != http.StatusOK {
+			t.Fatalf("expected status code 200, actual %d", w.Code)
+		}
+
+		checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+			filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+			filepath.Join(p, "blobs", "sha256-ca239d7bd8ea90e4a5d2e6bf88f8d74a47b14336e73eb4e18bed4dd325018116"),
+		})
+	})
+}

server/routes_delete_test.go

@@ -0,0 +1,107 @@
+package server
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"path/filepath"
+	"testing"
+
+	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
+	"github.com/ollama/ollama/types/model"
+)
+
+func TestDelete(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	envconfig.LoadConfig()
+
+	var s Server
+
+	w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test",
+		Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+		Name:      "test2",
+		Modelfile: fmt.Sprintf("FROM %s\nTEMPLATE {{ .System }} {{ .Prompt }}", createBinFile(t, nil, nil)),
+	})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test", "latest"),
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-8f2c2167d789c6b2302dff965160fa5029f6a24096d262c1cbb469f21a045382"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-ca239d7bd8ea90e4a5d2e6bf88f8d74a47b14336e73eb4e18bed4dd325018116"),
+		filepath.Join(p, "blobs", "sha256-fe7ac77b725cda2ccad03f88a880ecdfd7a33192d6cae08fce2c0ee1455991ed"),
+	})
+
+	w = createRequest(t, s.DeleteModelHandler, api.DeleteRequest{Name: "test"})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{
+		filepath.Join(p, "manifests", "registry.ollama.ai", "library", "test2", "latest"),
+	})
+
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
+		filepath.Join(p, "blobs", "sha256-8f2c2167d789c6b2302dff965160fa5029f6a24096d262c1cbb469f21a045382"),
+		filepath.Join(p, "blobs", "sha256-a4e5e156ddec27e286f75328784d7106b60a4eb1d246e950a001a3f944fbda99"),
+		filepath.Join(p, "blobs", "sha256-fe7ac77b725cda2ccad03f88a880ecdfd7a33192d6cae08fce2c0ee1455991ed"),
+	})
+
+	w = createRequest(t, s.DeleteModelHandler, api.DeleteRequest{Name: "test2"})
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{})
+	checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{})
+}
+
+func TestDeleteDuplicateLayers(t *testing.T) {
+	p := t.TempDir()
+	t.Setenv("OLLAMA_MODELS", p)
+	var s Server
+
+	n := model.ParseName("test")
+
+	var b bytes.Buffer
+	if err := json.NewEncoder(&b).Encode(&ConfigV2{}); err != nil {
+		t.Fatal(err)
+	}
+
+	config, err := NewLayer(&b, "application/vnd.docker.container.image.v1+json")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// create a manifest with duplicate layers
+	if err := WriteManifest(n, config, []*Layer{config}); err != nil {
+		t.Fatal(err)
+	}
+
+	w := createRequest(t, s.DeleteModelHandler, api.DeleteRequest{Name: "test"})
+	if w.Code != http.StatusOK {
+		t.Errorf("expected status code 200, actual %d", w.Code)
+	}
+
+	checkFileExists(t, filepath.Join(p, "manifests", "*", "*", "*", "*"), []string{})
+}

server/routes_list_test.go

@@ -0,0 +1,63 @@
+package server
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"slices"
+	"testing"
+
+	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
+)
+
+func TestList(t *testing.T) {
+	t.Setenv("OLLAMA_MODELS", t.TempDir())
+	envconfig.LoadConfig()
+
+	expectNames := []string{
+		"mistral:7b-instruct-q4_0",
+		"zephyr:7b-beta-q5_K_M",
+		"apple/OpenELM:latest",
+		"boreas:2b-code-v1.5-q6_K",
+		"notus:7b-v1-IQ2_S",
+		// TODO: host:port currently fails on windows (#4107)
+		// "localhost:5000/library/eurus:700b-v0.5-iq3_XXS",
+		"mynamespace/apeliotes:latest",
+		"myhost/mynamespace/lips:code",
+	}
+
+	var s Server
+	for _, n := range expectNames {
+		createRequest(t, s.CreateModelHandler, api.CreateRequest{
+			Name:      n,
+			Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+		})
+	}
+
+	w := createRequest(t, s.ListModelsHandler, nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status code 200, actual %d", w.Code)
+	}
+
+	var resp api.ListResponse
+	if err := json.NewDecoder(w.Body).Decode(&resp); err != nil {
+		t.Fatal(err)
+	}
+
+	if len(resp.Models) != len(expectNames) {
+		t.Fatalf("expected %d models, actual %d", len(expectNames), len(resp.Models))
+	}
+
+	actualNames := make([]string, len(resp.Models))
+	for i, m := range resp.Models {
+		actualNames[i] = m.Name
+	}
+
+	slices.Sort(actualNames)
+	slices.Sort(expectNames)
+
+	if !slices.Equal(actualNames, expectNames) {
+		t.Fatalf("expected slices to be equal %v", actualNames)
+	}
+}

server/routes_test.go

@@ -15,12 +15,37 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/parser"
+	"github.com/ollama/ollama/types/model"
 	"github.com/ollama/ollama/version"
 )
 
+func createTestFile(t *testing.T, name string) string {
+	t.Helper()
+
+	f, err := os.CreateTemp(t.TempDir(), name)
+	require.NoError(t, err)
+	defer f.Close()
+
+	err = binary.Write(f, binary.LittleEndian, []byte("GGUF"))
+	require.NoError(t, err)
+
+	err = binary.Write(f, binary.LittleEndian, uint32(3))
+	require.NoError(t, err)
+
+	err = binary.Write(f, binary.LittleEndian, uint64(0))
+	require.NoError(t, err)
+
+	err = binary.Write(f, binary.LittleEndian, uint64(0))
+	require.NoError(t, err)
+
+	return f.Name()
+}
+
 func Test_Routes(t *testing.T) {
 	type testCase struct {
 		Name     string
@@ -30,39 +55,19 @@ func Test_Routes(t *testing.T) {
 		Expected func(t *testing.T, resp *http.Response)
 	}
 
-	createTestFile := func(t *testing.T, name string) string {
+	createTestModel := func(t *testing.T, name string) {
 		t.Helper()
 
-		f, err := os.CreateTemp(t.TempDir(), name)
-		assert.Nil(t, err)
-		defer f.Close()
-
-		err = binary.Write(f, binary.LittleEndian, []byte("GGUF"))
-		assert.Nil(t, err)
-
-		err = binary.Write(f, binary.LittleEndian, uint32(3))
-		assert.Nil(t, err)
-
-		err = binary.Write(f, binary.LittleEndian, uint64(0))
-		assert.Nil(t, err)
-
-		err = binary.Write(f, binary.LittleEndian, uint64(0))
-		assert.Nil(t, err)
-
-		return f.Name()
-	}
-
-	createTestModel := func(t *testing.T, name string) {
 		fname := createTestFile(t, "ollama-model")
 
 		r := strings.NewReader(fmt.Sprintf("FROM %s\nPARAMETER seed 42\nPARAMETER top_p 0.9\nPARAMETER stop foo\nPARAMETER stop bar", fname))
 		modelfile, err := parser.ParseFile(r)
-		assert.Nil(t, err)
+		require.NoError(t, err)
 		fn := func(resp api.ProgressResponse) {
 			t.Logf("Status: %s", resp.Status)
 		}
-		err = CreateModel(context.TODO(), name, "", "", modelfile, fn)
-		assert.Nil(t, err)
+		err = CreateModel(context.TODO(), model.ParseName(name), "", "", modelfile, fn)
+		require.NoError(t, err)
 	}
 
 	testCases := []testCase{
@@ -74,9 +79,9 @@ func Test_Routes(t *testing.T) {
 			},
 			Expected: func(t *testing.T, resp *http.Response) {
 				contentType := resp.Header.Get("Content-Type")
-				assert.Equal(t, contentType, "application/json; charset=utf-8")
+				assert.Equal(t, "application/json; charset=utf-8", contentType)
 				body, err := io.ReadAll(resp.Body)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 				assert.Equal(t, fmt.Sprintf(`{"version":"%s"}`, version.Version), string(body))
 			},
 		},
@@ -86,17 +91,17 @@ func Test_Routes(t *testing.T) {
 			Path:   "/api/tags",
 			Expected: func(t *testing.T, resp *http.Response) {
 				contentType := resp.Header.Get("Content-Type")
-				assert.Equal(t, contentType, "application/json; charset=utf-8")
+				assert.Equal(t, "application/json; charset=utf-8", contentType)
 				body, err := io.ReadAll(resp.Body)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				var modelList api.ListResponse
 
 				err = json.Unmarshal(body, &modelList)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				assert.NotNil(t, modelList.Models)
-				assert.Equal(t, 0, len(modelList.Models))
+				assert.Empty(t, len(modelList.Models))
 			},
 		},
 		{
@@ -108,16 +113,18 @@ func Test_Routes(t *testing.T) {
 			},
 			Expected: func(t *testing.T, resp *http.Response) {
 				contentType := resp.Header.Get("Content-Type")
-				assert.Equal(t, contentType, "application/json; charset=utf-8")
+				assert.Equal(t, "application/json; charset=utf-8", contentType)
 				body, err := io.ReadAll(resp.Body)
-				assert.Nil(t, err)
+				require.NoError(t, err)
+
+				assert.NotContains(t, string(body), "expires_at")
 
 				var modelList api.ListResponse
 				err = json.Unmarshal(body, &modelList)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
-				assert.Equal(t, 1, len(modelList.Models))
-				assert.Equal(t, modelList.Models[0].Name, "test-model:latest")
+				assert.Len(t, modelList.Models, 1)
+				assert.Equal(t, "test-model:latest", modelList.Models[0].Name)
 			},
 		},
 		{
@@ -134,7 +141,7 @@ func Test_Routes(t *testing.T) {
 					Stream:    &stream,
 				}
 				jsonData, err := json.Marshal(createReq)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				req.Body = io.NopCloser(bytes.NewReader(jsonData))
 			},
@@ -142,11 +149,11 @@ func Test_Routes(t *testing.T) {
 				contentType := resp.Header.Get("Content-Type")
 				assert.Equal(t, "application/json", contentType)
 				_, err := io.ReadAll(resp.Body)
-				assert.Nil(t, err)
-				assert.Equal(t, resp.StatusCode, 200)
+				require.NoError(t, err)
+				assert.Equal(t, 200, resp.StatusCode)
 
 				model, err := GetModel("t-bone")
-				assert.Nil(t, err)
+				require.NoError(t, err)
 				assert.Equal(t, "t-bone:latest", model.ShortName)
 			},
 		},
@@ -161,13 +168,13 @@ func Test_Routes(t *testing.T) {
 					Destination: "beefsteak",
 				}
 				jsonData, err := json.Marshal(copyReq)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				req.Body = io.NopCloser(bytes.NewReader(jsonData))
 			},
 			Expected: func(t *testing.T, resp *http.Response) {
 				model, err := GetModel("beefsteak")
-				assert.Nil(t, err)
+				require.NoError(t, err)
 				assert.Equal(t, "beefsteak:latest", model.ShortName)
 			},
 		},
@@ -179,18 +186,18 @@ func Test_Routes(t *testing.T) {
 				createTestModel(t, "show-model")
 				showReq := api.ShowRequest{Model: "show-model"}
 				jsonData, err := json.Marshal(showReq)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 				req.Body = io.NopCloser(bytes.NewReader(jsonData))
 			},
 			Expected: func(t *testing.T, resp *http.Response) {
 				contentType := resp.Header.Get("Content-Type")
-				assert.Equal(t, contentType, "application/json; charset=utf-8")
+				assert.Equal(t, "application/json; charset=utf-8", contentType)
 				body, err := io.ReadAll(resp.Body)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				var showResp api.ShowResponse
 				err = json.Unmarshal(body, &showResp)
-				assert.Nil(t, err)
+				require.NoError(t, err)
 
 				var params []string
 				paramsSplit := strings.Split(showResp.Parameters, "\n")
@@ -210,6 +217,7 @@ func Test_Routes(t *testing.T) {
 	}
 
 	t.Setenv("OLLAMA_MODELS", t.TempDir())
+	envconfig.LoadConfig()
 
 	s := &Server{}
 	router := s.GenerateRoutes()
@@ -221,14 +229,14 @@ func Test_Routes(t *testing.T) {
 		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)
+			require.NoError(t, err)
 
 			if tc.Setup != nil {
 				tc.Setup(t, req)
 			}
 
 			resp, err := httpSrv.Client().Do(req)
-			assert.Nil(t, err)
+			require.NoError(t, err)
 			defer resp.Body.Close()
 
 			if tc.Expected != nil {
@@ -237,3 +245,83 @@ func Test_Routes(t *testing.T) {
 		})
 	}
 }
+
+func TestCase(t *testing.T) {
+	t.Setenv("OLLAMA_MODELS", t.TempDir())
+	envconfig.LoadConfig()
+
+	cases := []string{
+		"mistral",
+		"llama3:latest",
+		"library/phi3:q4_0",
+		"registry.ollama.ai/library/gemma:q5_K_M",
+		// TODO: host:port currently fails on windows (#4107)
+		// "localhost:5000/alice/bob:latest",
+	}
+
+	var s Server
+	for _, tt := range cases {
+		t.Run(tt, func(t *testing.T) {
+			w := createRequest(t, s.CreateModelHandler, api.CreateRequest{
+				Name:      tt,
+				Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+				Stream:    &stream,
+			})
+
+			if w.Code != http.StatusOK {
+				t.Fatalf("expected status 200 got %d", w.Code)
+			}
+
+			expect, err := json.Marshal(map[string]string{"error": "a model with that name already exists"})
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			t.Run("create", func(t *testing.T) {
+				w = createRequest(t, s.CreateModelHandler, api.CreateRequest{
+					Name:      strings.ToUpper(tt),
+					Modelfile: fmt.Sprintf("FROM %s", createBinFile(t, nil, nil)),
+					Stream:    &stream,
+				})
+
+				if w.Code != http.StatusBadRequest {
+					t.Fatalf("expected status 500 got %d", w.Code)
+				}
+
+				if !bytes.Equal(w.Body.Bytes(), expect) {
+					t.Fatalf("expected error %s got %s", expect, w.Body.String())
+				}
+			})
+
+			t.Run("pull", func(t *testing.T) {
+				w := createRequest(t, s.PullModelHandler, api.PullRequest{
+					Name:   strings.ToUpper(tt),
+					Stream: &stream,
+				})
+
+				if w.Code != http.StatusBadRequest {
+					t.Fatalf("expected status 500 got %d", w.Code)
+				}
+
+				if !bytes.Equal(w.Body.Bytes(), expect) {
+					t.Fatalf("expected error %s got %s", expect, w.Body.String())
+				}
+			})
+
+			t.Run("copy", func(t *testing.T) {
+				w := createRequest(t, s.CopyModelHandler, api.CopyRequest{
+					Source:      tt,
+					Destination: strings.ToUpper(tt),
+				})
+
+				if w.Code != http.StatusBadRequest {
+					t.Fatalf("expected status 500 got %d", w.Code)
+				}
+
+				if !bytes.Equal(w.Body.Bytes(), expect) {
+					t.Fatalf("expected error %s got %s", expect, w.Body.String())
+				}
+			})
+		})
+	}
+}

server/sched.go

@@ -7,17 +7,17 @@ import (
 	"log/slog"
 	"reflect"
 	"runtime"
+	"slices"
 	"sort"
 	"strings"
 	"sync"
 	"time"
 
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/gpu"
 	"github.com/ollama/ollama/llm"
-	"github.com/ollama/ollama/envconfig"
-	"golang.org/x/exp/slices"
 )
 
 type LlmRequest struct {
@@ -66,7 +66,7 @@ func (s *Scheduler) GetRunner(c context.Context, model *Model, opts api.Options,
 		opts.NumCtx = 4
 	}
 
-	opts.NumCtx = opts.NumCtx * envconfig.NumParallel
+	opts.NumCtx *= envconfig.NumParallel
 
 	req := &LlmRequest{
 		ctx:             c,
@@ -370,7 +370,6 @@ func (s *Scheduler) updateFreeSpace(allGpus gpu.GpuInfoList) {
 		r.refMu.Lock()
 		gpuIDs := make([]string, 0, len(r.gpus))
 		if r.llama != nil {
-
 			// TODO this should be broken down by GPU instead of assuming uniform spread
 			estimatedVRAMPerGPU := r.llama.EstimatedVRAM() / uint64(len(r.gpus))
 			for _, gpu := range r.gpus {
@@ -529,7 +528,6 @@ func (runner *runnerRef) waitForVRAMRecovery() chan interface{} {
 		}
 	}()
 	return finished
-
 }
 
 type ByDuration []*runnerRef

server/sched_test.go

@@ -12,11 +12,10 @@ import (
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/app/lifecycle"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/gpu"
 	"github.com/ollama/ollama/llm"
-	"github.com/ollama/ollama/envconfig"
-	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
@@ -53,10 +52,10 @@ func TestLoad(t *testing.T) {
 	}
 	gpus := gpu.GpuInfoList{}
 	s.load(req, ggml, gpus)
-	require.Len(t, req.successCh, 0)
+	require.Empty(t, req.successCh)
 	require.Len(t, req.errCh, 1)
 	s.loadedMu.Lock()
-	require.Len(t, s.loaded, 0)
+	require.Empty(t, s.loaded)
 	s.loadedMu.Unlock()
 	err := <-req.errCh
 	require.Contains(t, err.Error(), "this model may be incompatible")
@@ -113,7 +112,7 @@ func newScenario(t *testing.T, ctx context.Context, modelName string, estimatedV
 	t.Helper()
 
 	f, err := os.CreateTemp(t.TempDir(), modelName)
-	assert.Nil(t, err)
+	require.NoError(t, err)
 	defer f.Close()
 
 	gguf := llm.NewGGUFV3(binary.LittleEndian)
@@ -131,7 +130,7 @@ func newScenario(t *testing.T, ctx context.Context, modelName string, estimatedV
 	}, []llm.Tensor{
 		{Name: "blk.0.attn.weight", Kind: uint32(0), Offset: uint64(0), Shape: []uint64{1, 1, 1, 1}, WriterTo: &bytes.Reader{}},
 	})
-	assert.Nil(t, err)
+	require.NoError(t, err)
 
 	fname := f.Name()
 	model := &Model{Name: modelName, ModelPath: fname}
@@ -190,8 +189,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario1a.req.successCh:
 		require.Equal(t, resp.llama, scenario1a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario1a.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario1a.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -203,8 +202,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario1b.req.successCh:
 		require.Equal(t, resp.llama, scenario1a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario1b.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario1b.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -221,8 +220,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario2a.req.successCh:
 		require.Equal(t, resp.llama, scenario2a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario2a.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario2a.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -237,8 +236,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario3a.req.successCh:
 		require.Equal(t, resp.llama, scenario3a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario3a.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario3a.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -253,8 +252,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario3b.req.successCh:
 		require.Equal(t, resp.llama, scenario3b.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario3b.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario3b.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -269,8 +268,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario3c.req.successCh:
 		require.Equal(t, resp.llama, scenario3c.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario3c.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario3c.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -296,8 +295,8 @@ func TestRequests(t *testing.T) {
 	select {
 	case resp := <-scenario3d.req.successCh:
 		require.Equal(t, resp.llama, scenario3d.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, scenario3d.req.errCh, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, scenario3d.req.errCh)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -332,7 +331,7 @@ func TestGetRunner(t *testing.T) {
 	slog.Info("scenario1b")
 	successCh1b, errCh1b := s.GetRunner(scenario1b.ctx, scenario1b.req.model, scenario1b.req.opts, scenario1b.req.sessionDuration)
 	require.Len(t, s.pendingReqCh, 1)
-	require.Len(t, successCh1b, 0)
+	require.Empty(t, successCh1b)
 	require.Len(t, errCh1b, 1)
 	err := <-errCh1b
 	require.Contains(t, err.Error(), "server busy")
@@ -340,8 +339,8 @@ func TestGetRunner(t *testing.T) {
 	select {
 	case resp := <-successCh1a:
 		require.Equal(t, resp.llama, scenario1a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, errCh1a, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, errCh1a)
 	case <-ctx.Done():
 		t.Errorf("timeout")
 	}
@@ -355,9 +354,9 @@ func TestGetRunner(t *testing.T) {
 	successCh1c, errCh1c := s.GetRunner(scenario1c.ctx, scenario1c.req.model, scenario1c.req.opts, scenario1c.req.sessionDuration)
 	// Starts in pending channel, then should be quickly processsed to return an error
 	time.Sleep(5 * time.Millisecond)
-	require.Len(t, successCh1c, 0)
+	require.Empty(t, successCh1c)
 	s.loadedMu.Lock()
-	require.Len(t, s.loaded, 0)
+	require.Empty(t, s.loaded)
 	s.loadedMu.Unlock()
 	require.Len(t, errCh1c, 1)
 	err = <-errCh1c
@@ -386,8 +385,8 @@ func TestPrematureExpired(t *testing.T) {
 	select {
 	case resp := <-successCh1a:
 		require.Equal(t, resp.llama, scenario1a.srv)
-		require.Len(t, s.pendingReqCh, 0)
-		require.Len(t, errCh1a, 0)
+		require.Empty(t, s.pendingReqCh)
+		require.Empty(t, errCh1a)
 		s.loadedMu.Lock()
 		require.Len(t, s.loaded, 1)
 		s.loadedMu.Unlock()
@@ -401,9 +400,9 @@ func TestPrematureExpired(t *testing.T) {
 	time.Sleep(20 * time.Millisecond)
 	require.LessOrEqual(t, len(s.finishedReqCh), 1)
 	time.Sleep(10 * time.Millisecond)
-	require.Len(t, s.finishedReqCh, 0)
+	require.Empty(t, s.finishedReqCh)
 	s.loadedMu.Lock()
-	require.Len(t, s.loaded, 0)
+	require.Empty(t, s.loaded)
 	s.loadedMu.Unlock()
 
 	// also shouldn't happen in real life
@@ -487,7 +486,6 @@ func TestFindRunnerToUnload(t *testing.T) {
 	r2.refCount = 1
 	resp = s.findRunnerToUnload()
 	require.Equal(t, r1, resp)
-
 }
 
 func TestNeedsReload(t *testing.T) {

server/upload.go

@@ -146,7 +146,7 @@ func (b *blobUpload) Run(ctx context.Context, opts *registryOptions) {
 		case requestURL := <-b.nextURL:
 			g.Go(func() error {
 				var err error
-				for try := 0; try < maxRetries; try++ {
+				for try := range maxRetries {
 					err = b.uploadPart(inner, http.MethodPatch, requestURL, part, opts)
 					switch {
 					case errors.Is(err, context.Canceled):
@@ -190,7 +190,7 @@ func (b *blobUpload) Run(ctx context.Context, opts *registryOptions) {
 	headers.Set("Content-Type", "application/octet-stream")
 	headers.Set("Content-Length", "0")
 
-	for try := 0; try < maxRetries; try++ {
+	for try := range maxRetries {
 		var resp *http.Response
 		resp, err = makeRequestWithRetry(ctx, http.MethodPut, requestURL, headers, nil, opts)
 		if errors.Is(err, context.Canceled) {
@@ -253,7 +253,7 @@ func (b *blobUpload) uploadPart(ctx context.Context, method string, requestURL *
 		}
 
 		// retry uploading to the redirect URL
-		for try := 0; try < maxRetries; try++ {
+		for try := range maxRetries {
 			err = b.uploadPart(ctx, http.MethodPut, redirectURL, part, nil)
 			switch {
 			case errors.Is(err, context.Canceled):
@@ -391,7 +391,7 @@ func uploadBlob(ctx context.Context, mp ModelPath, layer *Layer, opts *registryO
 			return err
 		}
 
-		// nolint: contextcheck
+		//nolint:contextcheck
 		go upload.Run(context.Background(), opts)
 	}
 

templates/alfred.gotmpl

@@ -0,0 +1 @@
+{{ if .System }}<start_system>{{ .System }}<end_message>{{ end }}{{ if .Prompt }}<start_user>{{ .Prompt }}<end_message>{{ end }}<start_assistant>{{ .Response }}<end_message>

templates/alpaca.gotmpl

@@ -0,0 +1,7 @@
+{{ if .System }}{{ .System }}
+
+{{ end }}{{ if .Prompt }}### Instruction:
+{{ .Prompt }}
+
+{{ end }}### Response:
+{{ .Response }}

templates/chatml.gotmpl

@@ -0,0 +1,6 @@
+{{ if .System }}<|im_start|>system
+{{ .System }}<|im_end|>
+{{ end }}{{ if .Prompt }}<|im_start|>user
+{{ .Prompt }}<|im_end|>
+{{ end }}<|im_start|>assistant
+{{ .Response }}<|im_end|>

templates/chatqa.gotmpl

@@ -0,0 +1,5 @@
+{{ if .System }}System: {{ .System }}
+
+{{ end }}{{ if .Prompt }}User: {{ .Prompt }}
+
+{{ end }}Assistant: <|begin_of_text|>{{ .Response }}

templates/codellama-70b-instruct.gotmpl

@@ -0,0 +1,8 @@
+{{ if .System }} Source: system
+
+ {{ .System }} <step>{{ end }} Source: user
+
+ {{ .Prompt }} <step> Source: assistant
+Destination: user
+
+ {{ .Response }}<step>

templates/falcon-instruct.gotmpl

@@ -0,0 +1,3 @@
+{{ if .System }}{{ .System }}
+{{ end }}{{ if .Prompt }}User: {{ .Prompt }}
+{{ end }}Assistant: {{ .Response }}

templates/gemma-instruct.gotmpl

@@ -0,0 +1,4 @@
+<start_of_turn>user
+{{ if .System }}{{ .System }} {{ end }}{{ .Prompt }}<end_of_turn>
+<start_of_turn>model
+{{ .Response }}<end_of_turn>

templates/granite-instruct.gotmpl

@@ -0,0 +1,9 @@
+{{ if .System }}
+System:
+{{ .System }}
+
+{{ end }}{{ if .Prompt }}Question:
+{{ .Prompt }}
+
+{{ end }}Answer:
+{{ .Response }}

templates/index.json

@@ -0,0 +1,138 @@
+[
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ system_message }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|im_start|>user\\n' + content + '<|im_end|>\\n<|im_start|>assistant\\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|im_end|>' + '\\n' }}{% endif %}{% endfor %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
+    "name": "zephyr"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'user' or messages[0]['role'] == 'system' %}{{ bos_token }}{% endif %}{% for message in messages %}{{ '<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n' }}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% elif messages[-1]['role'] == 'assistant' %}{{ eos_token }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{{ bos_token }}{% for message in messages %}{{ 'GPT4 Correct ' + message['role'].title() + ': ' + message['content'] + '<|end_of_turn|>'}}{% endfor %}{% if add_generation_prompt %}{{ 'GPT4 Correct Assistant:' }}{% endif %}",
+    "name": "openchat"
+  },
+  {
+    "template": "{{bos_token}}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
+    "name": "zephyr"
+  },
+  {
+    "template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",
+    "name": "mistral-instruct"
+  },
+  {
+    "template": "{{bos_token}}{{'You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.\n\n'}}\n{%- for message in messages %}\n    {%- if message['role'] == 'system' %}\n        {{ raise_exception('System messages are not allowed in this template.') }}\n    {%- else %}\n        {%- if message['role'] == 'user' %}\n{{'### Instruction\n' + message['content'] + '\n\n'}}\n        {%- else %}\n{{'### Response\n' + message['content'] + eos_token + '\n\n'}}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{{'### Response\n'}}",
+    "name": "starcoder2-instruct"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 and system_message != false %}{% set content = '<<SYS>>\\n' + system_message + '\\n<</SYS>>\\n\\n' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ bos_token + '[INST] ' + content | trim + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' '  + content | trim + ' ' + eos_token }}{% endif %}{% endfor %}",
+    "name": "llama2-chat"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set user_index = 1 %}{% else %}{% set user_index = 0 %}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != ((loop.index0 + user_index) % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 %}{{ '<s>' }}{% endif %}{% set content = 'Source: ' + message['role'] + '\n\n ' + message['content'] | trim %}{{ content + ' <step> ' }}{% endfor %}{{'Source: assistant\nDestination: user\n\n '}}",
+    "name": "codellama-70b-instruct"
+  },
+  {
+    "template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token + ' ' }}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",
+    "name": "mistral-instruct"
+  },
+  {
+    "template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|im_start|>user\n' + message['content'] + '<|im_end|>' }}\n{% elif message['role'] == 'system' %}\n{{ '<|im_start|>system\n' + message['content'] + '<|im_end|>' }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|im_start|>assistant\n' + message['content'] + '<|im_end|>' }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|im_start|>assistant' }}\n{% endif %}\n{% endfor %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = 'You are a helpful assistant.' %}{% endif %}{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in loop_messages %}{% if loop.index0 == 0 %}{{'<|im_start|>system\n' + system_message + '<|im_end|>\n'}}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% elif 'system' not in messages[0]['role'] %}{% set loop_messages = messages %}{% set system_message = 'You are DBRX, created by Databricks. You were last updated in December 2023. You answer questions based on information available up to that point.\nYOU PROVIDE SHORT RESPONSES TO SHORT QUESTIONS OR STATEMENTS, but provide thorough responses to more complex and open-ended questions.\nYou assist with various tasks, from writing to coding (using markdown for code blocks \u2014 remember to use ``` with code, JSON, and tables).\n(You do not have real-time data access or code execution capabilities. You avoid stereotyping and provide balanced perspectives on controversial topics. You do not provide song lyrics, poems, or news articles and do not divulge details of your training data.)\nThis is your system prompt, guiding your responses. Do not reference it, just respond to the user. If you find yourself talking about this message, stop. You should be responding appropriately and usually that means not mentioning this.\nYOU DO NOT MENTION ANY OF THIS INFORMATION ABOUT YOURSELF UNLESS THE INFORMATION IS DIRECTLY PERTINENT TO THE USER\\'S QUERY.' %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% for message in loop_messages %}{% if loop.index0 == 0 %}{% if system_message != false %}{{ '<|im_start|>system\n' + system_message | trim + '<|im_end|>\n'}}{% endif %}{{ '<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' }}{% else %}{{ '\n' + '<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' }}{% endif %}{% if (add_generation_prompt == true and loop.last) %}{{ '\n' + '<|im_start|>' + 'assistant' + '\n' }}{% endif %}{% endfor %}",
+    "name": "chatml"
+  },
+  {
+    "template": "{% if not add_generation_prompt is defined %}\n{% set add_generation_prompt = false %}\n{% endif %}\n{%- set ns = namespace(found=false) -%}\n{%- for message in messages -%}\n    {%- if message['role'] == 'system' -%}\n        {%- set ns.found = true -%}\n    {%- endif -%}\n{%- endfor -%}\n{{bos_token}}{%- if not ns.found -%}\n{{'You are an AI programming assistant, utilizing the Deepseek Coder model, developed by Deepseek Company, and you only answer questions related to computer science. For politically sensitive questions, security and privacy issues, and other non-computer science questions, you will refuse to answer\\n'}}\n{%- endif %}\n{%- for message in messages %}\n    {%- if message['role'] == 'system' %}\n{{ message['content'] }}\n    {%- else %}\n        {%- if message['role'] == 'user' %}\n{{'### Instruction:\\n' + message['content'] + '\\n'}}\n        {%- else %}\n{{'### Response:\\n' + message['content'] + '\\n<|EOT|>\\n'}}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{% if add_generation_prompt %}\n{{'### Response:'}}\n{% endif %}",
+    "name": "alpaca"
+  },
+  {
+    "template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{% if message['role'] == 'user' %}{{ 'User: ' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Assistant: ' + message['content'] + eos_token }}{% elif message['role'] == 'system' %}{{ message['content'] + '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
+    "name": "chatqa"
+  },
+  {
+    "template": "{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '<start_of_turn>' + role + '\n' + message['content'] | trim + '<end_of_turn>\n' }}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model\n'}}{% endif %}",
+    "name": "gemma-instruct"
+  },
+  {
+    "template": "{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}{% endif %}",
+    "name": "llama3-instruct"
+  },
+  {
+    "template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ 'Question:\n' + message['content'] + '\n\n' }}{% elif message['role'] == 'system' %}\n{{ 'System:\n' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Answer:\n'  + message['content'] + '\n\n' }}{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ 'Answer:\n' }}{% endif %}{% endfor %}",
+    "name": "granite-instruct"
+  },
+  {
+    "template": "{{bos_token}}{{'You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.\n\n'}}\n{%- for message in messages %}\n    {%- if message['role'] == 'system' %}\n        {{ raise_exception('System messages are not allowed in this template.') }}\n    {%- else %}\n        {%- if message['role'] == 'user' %}\n{{'@@ Instruction\n' + message['content'] + '\n\n'}}\n        {%- else %}\n{{'@@ Response\n' + message['content'] + eos_token + '\n\n'}}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{{'@@ Response\n'}}",
+    "name": "magicoder"
+  },
+  {
+    "template": "{% for message in messages %}{% if message['role'] == 'user' %}{{ '<start_user>' + message['content'].strip() + '<end_message>' }}{% elif message['role'] == 'system' %}{{ '<start_system>' + message['content'].strip() + '<end_message>' }}{% elif message['role'] == 'assistant' %}{{ '<start_assistant>'  + message['content'] + '<end_message>' }}{% else %}{{ raise_exception('Only system, user and assistant roles are supported.') }}{% endif %}{% if loop.last and add_generation_prompt %}{{ '<start_assistant>' }}{% endif %}{% endfor %}",
+    "name": "alfred"
+  },
+  {
+    "template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 and system_message != false %}{% set content = '<<SYS>>\\n' + system_message + '\\n<</SYS>>\\n\\n' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ bos_token + '[INST] ' + content.strip() + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' '  + content.strip() + ' ' + eos_token }}{% endif %}{% endfor %}",
+    "name": "llama2-chat"
+  },
+  {
+    "template": "{% for message in messages %}{% if (message['role'] == 'user') %}{{'<|user|>' + '\n' + message['content'] + '<|end|>' + '\n' + '<|assistant|>' + '\n'}}{% elif (message['role'] == 'assistant') %}{{message['content'] + '<|end|>' + '\n'}}{% endif %}{% endfor %}",
+    "name": "phi-3"
+  },
+  {
+    "template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') %}{{'<|user|>' + '\n' + message['content'] + '<|end|>' + '\n' + '<|assistant|>' + '\n'}}{% elif (message['role'] == 'assistant') %}{{message['content'] + '<|end|>' + '\n'}}{% endif %}{% endfor %}",
+    "name": "phi-3"
+  },
+  {
+    "template": "{{ bos_token }}{% for message in messages %}{{'<|' + message['role'] + '|>' + '\n' + message['content'] + '<|end|>\n' }}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>\n' }}{% else %}{{ eos_token }}{% endif %}",
+    "name": "phi-3"
+  },
+  {
+    "template": "{{ bos_token }}{%- if messages[0]['role'] == 'system' -%}{% set loop_messages = messages[1:] %}{%- else -%}{% set loop_messages = messages %}{% endif %}System: This is a chat between a user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions based on the context. The assistant should also indicate when the answer cannot be found in the context.\n\n{% for message in loop_messages %}{%- if message['role'] == 'user' -%}User: {{ message['content'].strip() + '\n\n' }}{%- else -%}Assistant: {{ message['content'].strip() + '\n\n' }}{%- endif %}{% if loop.last and message['role'] == 'user' %}Assistant:{% endif %}{% endfor %}",
+    "name": "chatqa"
+  },
+  {
+    "template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ 'User: \n' + message['content'] }}\n{% elif message['role'] == 'system' %}\n{{ 'System: ' + message['content'] }}\n{% elif message['role'] == 'assistant' %}\n{{ 'Falcon:\n'  + message['content']}}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ 'Falcon:' }}\n{% endif %}\n{% endfor %}",
+    "name": "falcon-instruct"
+  },
+  {
+    "template": "{% for message in messages %}{% if not loop.first %}{{ '\n' }}{% endif %}{% if message['role'] == 'system' %}{{ 'System: ' }}{% elif message['role'] == 'user' %}{{ 'User: ' }}{% elif message['role'] == 'assistant' %}{{ 'Falcon: ' }}{% endif %}{{ message['content'] }}{% endfor %}{% if add_generation_prompt %}{{ '\n' + 'Falcon:' }}{% endif %}",
+    "name": "falcon-instruct"
+  },
+  {
+    "template": "{% for message in messages %}{% if message['role'] == 'system' %}{% if message['content']%}{{'### System:\n' + message['content']+'\n\n'}}{% endif %}{% elif message['role'] == 'user' %}{{'### User:\n' + message['content']+'\n\n'}}{% elif message['role'] == 'assistant' %}{{'### Assistant:\n'  + message['content']}}{% endif %}{% if loop.last and add_generation_prompt %}{{ '### Assistant:\n' }}{% endif %}{% endfor %}",
+    "name": "solar-instruct"
+  }
+]

templates/llama2-chat.gotmpl

@@ -0,0 +1,3 @@
+[INST] <<SYS>>{{ .System }}<</SYS>>
+
+{{ .Prompt }} [/INST] {{ .Response }}

templates/llama3-instruct.gotmpl

@@ -0,0 +1,7 @@
+{{ if .System }}<|start_header_id|>system<|end_header_id|>
+
+{{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>
+
+{{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>
+
+{{ .Response }}<|eot_id|>

templates/magicoder.gotmpl

@@ -0,0 +1,7 @@
+{{ if .System }}{{ .System }}
+
+{{ end }}{{ if .Prompt }}@@ Instruction
+{{ .Prompt }}
+
+{{ end }}@@ Response
+{{ .Response }}

templates/mistral-instruct.gotmpl

@@ -0,0 +1,6 @@
+{{ if .System }}<|im_start|>system
+{{ .System }}<|im_end|>
+{{ end }}{{ if .Prompt }}<|im_start|>user
+{{ .Prompt }}<|im_end|>
+{{ end }}<|im_start|>assistant
+{{ .Response }}<|im_end|>

templates/openchat.gotmpl

@@ -0,0 +1 @@
+{{ .System }}<|end_of_turn|>GPT4 Correct User: {{ .Prompt }}<|end_of_turn|>GPT4 Correct Assistant: {{ .Response }}<|end_of_turn|>

templates/phi-3.gotmpl

@@ -0,0 +1,6 @@
+{{ if .System }}<|system|>
+{{ .System }}<|end|>
+{{ end }}{{ if .Prompt }}<|user|>
+{{ .Prompt }}<|end|>
+{{ end }}<|assistant|>
+{{ .Response }}<|end|>

templates/solar-instruct.gotmpl

@@ -0,0 +1,8 @@
+{{ if .System }}### System:
+{{ .System }}
+
+{{ end }}{{ if .Prompt }}### User:
+{{ .Prompt }}
+
+{{ end }}### Assistant:
+{{ .Response }}

templates/starcoder2-instruct.gotmpl

@@ -0,0 +1,9 @@
+{{ if .System }}{{ .System }}
+
+{{ end }}{{ if .Prompt }}### Instruction
+{{ .Prompt }}
+
+
+{{ end }}### Response
+{{ .Response }}<|endoftext|>
+