Merge pull request #5032 from dhiltgen/actually_skip

Actually skip PhysX on windows

README.md

@@ -332,6 +332,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [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)

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,

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"
@@ -377,8 +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

cmd/cmd.go

@@ -960,17 +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
 	}

docs/api.md

@@ -250,7 +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
 
@@ -259,8 +259,7 @@ curl http://localhost:11434/api/generate -d '{
   "model": "mistral",
   "prompt": "Why is the sky blue?",
   "options": {
-    "seed": 123,
-    "temperature": 0
+    "seed": 123
   }
 }'
 ```

envconfig/config.go

@@ -1,6 +1,7 @@
 package envconfig
 
 import (
+	"errors"
 	"fmt"
 	"log/slog"
 	"net"
@@ -11,6 +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
@@ -18,6 +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
@@ -26,6 +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
@@ -50,13 +67,13 @@ 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_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, "Maximum VRAM"},
-		"OLLAMA_MODELS":            {"OLLAMA_MODELS", "", "The path to the models directory"},
+		"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)"},
@@ -216,4 +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))
+			}
+		})
+	}
+}

gpu/gpu.go

@@ -296,6 +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)

llm/ext_server/server.cpp

@@ -359,7 +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;
@@ -483,9 +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);
     }
 
@@ -584,7 +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);
@@ -811,7 +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;

llm/ggml.go

@@ -231,7 +231,8 @@ const (
 	// Magic constant for `ggla` files (LoRA adapter).
 	FILE_MAGIC_GGLA = 0x67676C61
 	// Magic constant for `gguf` files (versioned, gguf)
-	FILE_MAGIC_GGUF = 0x46554747
+	FILE_MAGIC_GGUF_LE = 0x46554747
+	FILE_MAGIC_GGUF_BE = 0x47475546
 )
 
 var ErrUnsupportedFormat = errors.New("unsupported model format")
@@ -246,7 +247,7 @@ func DetectGGMLType(b []byte) string {
 		return "ggjt"
 	case FILE_MAGIC_GGLA:
 		return "ggla"
-	case FILE_MAGIC_GGUF:
+	case FILE_MAGIC_GGUF_LE, FILE_MAGIC_GGUF_BE:
 		return "gguf"
 	default:
 		return ""
@@ -254,19 +255,21 @@ func DetectGGMLType(b []byte) string {
 }
 
 func DecodeGGML(rs io.ReadSeeker) (*GGML, int64, error) {
-	var magic [4]byte
+	var magic uint32
 	if err := binary.Read(rs, binary.LittleEndian, &magic); err != nil {
 		return nil, 0, err
 	}
 
 	var c container
-	switch binary.LittleEndian.Uint32(magic[:]) {
+	switch magic {
 	case FILE_MAGIC_GGML, FILE_MAGIC_GGMF, FILE_MAGIC_GGJT:
 		return nil, 0, ErrUnsupportedFormat
 	case FILE_MAGIC_GGLA:
 		c = &containerGGLA{}
-	case FILE_MAGIC_GGUF:
+	case FILE_MAGIC_GGUF_LE:
 		c = &containerGGUF{ByteOrder: binary.LittleEndian}
+	case FILE_MAGIC_GGUF_BE:
+		c = &containerGGUF{ByteOrder: binary.BigEndian}
 	default:
 		return nil, 0, errors.New("invalid file magic")
 	}

llm/gguf.go

@@ -36,23 +36,10 @@ func (c *containerGGUF) Name() string {
 }
 
 func (c *containerGGUF) Decode(rs io.ReadSeeker) (model, error) {
-	var version [4]byte
-	if err := binary.Read(rs, c.ByteOrder, &version); err != nil {
+	if err := binary.Read(rs, c.ByteOrder, &c.Version); err != nil {
 		return nil, err
 	}
 
-	// if the lower 16 bits are 0, the byte order is probably wrong
-	if c.ByteOrder.Uint32(version[:])&1<<4 == 0 {
-		switch c.ByteOrder {
-		case binary.LittleEndian:
-			c.ByteOrder = binary.BigEndian
-		case binary.BigEndian:
-			c.ByteOrder = binary.LittleEndian
-		}
-	}
-
-	c.Version = c.ByteOrder.Uint32(version[:])
-
 	var err error
 	switch c.Version {
 	case 1:

parser/parser.go

@@ -3,15 +3,14 @@ package parser
 import (
 	"bufio"
 	"bytes"
-	"encoding/binary"
 	"errors"
 	"fmt"
 	"io"
-	"log/slog"
 	"strconv"
 	"strings"
-	"unicode/utf16"
-	"unicode/utf8"
+
+	"golang.org/x/text/encoding/unicode"
+	"golang.org/x/text/transform"
 )
 
 type File struct {
@@ -74,26 +73,14 @@ func ParseFile(r io.Reader) (*File, error) {
 
 	var f File
 
-	br := bufio.NewReader(r)
+	tr := unicode.BOMOverride(unicode.UTF8.NewDecoder())
+	br := bufio.NewReader(transform.NewReader(r, tr))
 
-	var sc scannerDecoder = utf8ScannerDecoder{}
-	if bom, err := br.Peek(2); err != nil {
-		slog.Warn("error reading byte-order mark", "error", err)
-	} else if bytes.Equal(bom, []byte{0xFE, 0xFF}) {
-		sc = utf16ScannerDecoder{binary.LittleEndian}
-		//nolint:errcheck
-		br.Discard(2)
-	} else if bytes.Equal(bom, []byte{0xFF, 0xFE}) {
-		sc = utf16ScannerDecoder{binary.BigEndian}
-		//nolint:errcheck
-		br.Discard(2)
-	}
-
-	scanner := bufio.NewScanner(br)
-	scanner.Split(sc.ScanBytes)
-	for scanner.Scan() {
-		r, err := sc.DecodeRune(scanner.Bytes())
-		if err != nil {
+	for {
+		r, _, err := br.ReadRune()
+		if errors.Is(err, io.EOF) {
+			break
+		} else if err != nil {
 			return nil, err
 		}
 
@@ -315,39 +302,3 @@ func isValidCommand(cmd string) bool {
 		return false
 	}
 }
-
-type scannerDecoder interface {
-	ScanBytes(data []byte, atEOF bool) (advance int, token []byte, err error)
-	DecodeRune([]byte) (rune, error)
-}
-
-type utf8ScannerDecoder struct{}
-
-func (utf8ScannerDecoder) ScanBytes(data []byte, atEOF bool) (advance int, token []byte, err error) {
-	return scanBytesN(data, 1, atEOF)
-}
-
-func (utf8ScannerDecoder) DecodeRune(data []byte) (rune, error) {
-	r, _ := utf8.DecodeRune(data)
-	return r, nil
-}
-
-type utf16ScannerDecoder struct {
-	binary.ByteOrder
-}
-
-func (utf16ScannerDecoder) ScanBytes(data []byte, atEOF bool) (advance int, token []byte, err error) {
-	return scanBytesN(data, 2, atEOF)
-}
-
-func (e utf16ScannerDecoder) DecodeRune(data []byte) (rune, error) {
-	return utf16.Decode([]uint16{e.ByteOrder.Uint16(data)})[0], nil
-}
-
-func scanBytesN(data []byte, n int, atEOF bool) (int, []byte, error) {
-	if atEOF && len(data) == 0 {
-		return 0, nil, nil
-	}
-
-	return n, data[:n], nil
-}

parser/parser_test.go

@@ -11,6 +11,8 @@ import (
 
 	"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) {
@@ -517,14 +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)
-	require.NoError(t, err)
-
-	actual, err := ParseFile(buf)
-	require.NoError(t, err)
 
 	expected := []Command{
 		{Name: "model", Args: "bob"},
@@ -533,14 +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)
-	require.NoError(t, err)
+		actual, err := ParseFile(&b)
+		require.NoError(t, err)
 
-	actual, err = ParseFile(buf)
-	require.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)
+		})
+	}
 }

server/images.go

@@ -28,7 +28,6 @@ import (
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/llm"
 	"github.com/ollama/ollama/parser"
-	"github.com/ollama/ollama/templates"
 	"github.com/ollama/ollama/types/errtypes"
 	"github.com/ollama/ollama/types/model"
 	"github.com/ollama/ollama/version"
@@ -333,7 +332,7 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 
 		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 {
@@ -435,20 +434,6 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 					config.ModelType = cmp.Or(config.ModelType, format.HumanNumber(baseLayer.GGML.KV().ParameterCount()))
 					config.FileType = cmp.Or(config.FileType, baseLayer.GGML.KV().FileType().String())
 					config.ModelFamilies = append(config.ModelFamilies, baseLayer.GGML.KV().Architecture())
-
-					if s := baseLayer.GGML.KV().ChatTemplate(); s != "" {
-						if t, err := templates.NamedTemplate(s); err != nil {
-							slog.Debug("template detection", "error", err)
-						} else {
-							layer, err := NewLayer(t.Reader(), "application/vnd.ollama.image.template")
-							if err != nil {
-								return err
-							}
-
-							layer.status = fmt.Sprintf("using autodetected template %s", t.Name)
-							layers = append(layers, layer)
-						}
-					}
 				}
 
 				layers = append(layers, baseLayer.Layer)
@@ -975,7 +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 ""
 	}
 

server/manifest_test.go

@@ -7,6 +7,7 @@ import (
 	"slices"
 	"testing"
 
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/types/model"
 )
 
@@ -107,6 +108,7 @@ func TestManifests(t *testing.T) {
 		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)

server/model.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
@@ -14,17 +15,18 @@ 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) {
+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):
@@ -66,16 +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
@@ -179,13 +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 {
@@ -227,10 +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

@@ -7,6 +7,8 @@ import (
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+
+	"github.com/ollama/ollama/envconfig"
 )
 
 func TestGetBlobsPath(t *testing.T) {
@@ -61,6 +63,7 @@ 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)
 

server/routes_create_test.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/llm"
 )
 
@@ -86,6 +87,7 @@ func checkFileExists(t *testing.T, p string, expect []string) {
 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{
@@ -111,6 +113,7 @@ func TestCreateFromBin(t *testing.T) {
 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{
@@ -151,6 +154,7 @@ func TestCreateFromModel(t *testing.T) {
 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{
@@ -197,6 +201,7 @@ func TestCreateRemovesLayers(t *testing.T) {
 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{
@@ -252,6 +257,7 @@ func TestCreateUnsetsSystem(t *testing.T) {
 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{
@@ -354,6 +360,7 @@ func TestCreateMergeParameters(t *testing.T) {
 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{
@@ -429,6 +436,7 @@ func TestCreateReplacesMessages(t *testing.T) {
 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{
@@ -474,6 +482,7 @@ func TestCreateTemplateSystem(t *testing.T) {
 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{
@@ -519,6 +528,7 @@ func TestCreateLicenses(t *testing.T) {
 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) {
@@ -535,7 +545,7 @@ func TestCreateDetectTemplate(t *testing.T) {
 		}
 
 		checkFileExists(t, filepath.Join(p, "blobs", "*"), []string{
-			filepath.Join(p, "blobs", "sha256-06cd2687a518d624073f125f1db1c5c727f77c75e84a138fe745186dbbbb4cd7"),
+			filepath.Join(p, "blobs", "sha256-2f8e594e6f34b1b4d36a246628eeb3365ce442303d656f1fcc69e821722acea0"),
 			filepath.Join(p, "blobs", "sha256-542b217f179c7825eeb5bca3c77d2b75ed05bafbd3451d9188891a60a85337c6"),
 			filepath.Join(p, "blobs", "sha256-553c4a3f747b3d22a4946875f1cc8ed011c2930d83f864a0c7265f9ec0a20413"),
 		})

server/routes_delete_test.go

@@ -9,12 +9,15 @@ import (
 	"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{

server/routes_list_test.go

@@ -8,10 +8,12 @@ import (
 	"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",

server/routes_test.go

@@ -18,6 +18,7 @@ import (
 	"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"
@@ -216,6 +217,7 @@ func Test_Routes(t *testing.T) {
 	}
 
 	t.Setenv("OLLAMA_MODELS", t.TempDir())
+	envconfig.LoadConfig()
 
 	s := &Server{}
 	router := s.GenerateRoutes()
@@ -246,6 +248,7 @@ func Test_Routes(t *testing.T) {
 
 func TestCase(t *testing.T) {
 	t.Setenv("OLLAMA_MODELS", t.TempDir())
+	envconfig.LoadConfig()
 
 	cases := []string{
 		"mistral",