return 204 for cross-origin OPTIONS

* simplify input output

* direct comp

* in line image

* rm error pointer type

* update response testing

* lint

.gitattributes

@@ -1,2 +1,3 @@
 llm/ext_server/* linguist-vendored
-* text eol=lf
+* text=auto
+*.go text eol=lf

README.md

@@ -325,6 +325,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [tlm](https://github.com/yusufcanb/tlm)
 - [podman-ollama](https://github.com/ericcurtin/podman-ollama)
 - [gollama](https://github.com/sammcj/gollama)
+- [Ollama eBook Summary](https://github.com/cognitivetech/ollama-ebook-summary/)
 
 ### Database
 

docs/api.md

@@ -669,7 +669,7 @@ curl http://localhost:11434/api/chat -d '{
 
 ```
 curl http://localhost:11434/api/chat -d '{
-  "model": "mistral",
+  "model": "llama3.1",
   "messages": [
     {
       "role": "user",
@@ -708,7 +708,7 @@ curl http://localhost:11434/api/chat -d '{
 
 ```json
 {
-  "model": "mistral:7b-instruct-v0.3-q4_K_M",
+  "model": "llama3.1",
   "created_at": "2024-07-22T20:33:28.123648Z",
   "message": {
     "role": "assistant",
@@ -1175,7 +1175,10 @@ curl http://localhost:11434/api/embed -d '{
   "embeddings": [[
     0.010071029, -0.0017594862, 0.05007221, 0.04692972, 0.054916814,
     0.008599704, 0.105441414, -0.025878139, 0.12958129, 0.031952348
-  ]]
+  ]],
+  "total_duration": 14143917,
+  "load_duration": 1019500,
+  "prompt_eval_count": 8
 }
 ```
 

gpu/assets.go

@@ -49,13 +49,9 @@ func PayloadsDir() (string, error) {
 		}
 
 		// Track our pid so we can clean up orphaned tmpdirs
-		pidFilePath := filepath.Join(tmpDir, "ollama.pid")
-		pidFile, err := os.OpenFile(pidFilePath, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, os.ModePerm)
-		if err != nil {
-			return "", err
-		}
-		if _, err := pidFile.Write([]byte(strconv.Itoa(os.Getpid()))); err != nil {
-			return "", err
+		n := filepath.Join(tmpDir, "ollama.pid")
+		if err := os.WriteFile(n, []byte(strconv.Itoa(os.Getpid())), 0o644); err != nil {
+			return "", fmt.Errorf("failed to write pid file %s: %w", n, err)
 		}
 
 		// We create a distinct subdirectory for payloads within the tmpdir
@@ -67,37 +63,44 @@ func PayloadsDir() (string, error) {
 
 // Best effort to clean up prior tmpdirs
 func cleanupTmpDirs() {
-	dirs, err := filepath.Glob(filepath.Join(os.TempDir(), "ollama*"))
+	matches, err := filepath.Glob(filepath.Join(os.TempDir(), "ollama*", "ollama.pid"))
 	if err != nil {
 		return
 	}
-	for _, d := range dirs {
-		info, err := os.Stat(d)
-		if err != nil || !info.IsDir() {
+
+	for _, match := range matches {
+		raw, err := os.ReadFile(match)
+		if errors.Is(err, os.ErrNotExist) {
+			slog.Debug("not a ollama runtime directory, skipping", "path", match)
 			continue
-		}
-		raw, err := os.ReadFile(filepath.Join(d, "ollama.pid"))
-		if err != nil {
-			slog.Warn("failed to read ollama.pid", "path", d, "error", err)
-			// No pid, ignore this tmpdir
+		} else if err != nil {
+			slog.Warn("could not read ollama.pid, skipping", "path", match, "error", err)
 			continue
 		}
 
 		pid, err := strconv.Atoi(string(raw))
 		if err != nil {
-			slog.Warn("failed to parse pid", "path", d, "error", err)
+			slog.Warn("invalid pid, skipping", "path", match, "error", err)
 			continue
 		}
 
-		proc, err := os.FindProcess(pid)
-		if err == nil && !errors.Is(proc.Signal(syscall.Signal(0)), os.ErrProcessDone) {
-			slog.Warn("found running ollama", "pid", pid, "path", d)
-			// Another running ollama, ignore this tmpdir
+		p, err := os.FindProcess(pid)
+		if err == nil && !errors.Is(p.Signal(syscall.Signal(0)), os.ErrProcessDone) {
+			slog.Warn("process still running, skipping", "pid", pid, "path", match)
 			continue
 		}
 
-		if err := os.Remove(d); err != nil {
-			slog.Warn("unable to cleanup stale tmpdir", "path", d, "error", err)
+		if err := os.Remove(match); err != nil {
+			slog.Warn("could not cleanup stale pidfile", "path", match, "error", err)
+		}
+
+		runners := filepath.Join(filepath.Dir(match), "runners")
+		if err := os.RemoveAll(runners); err != nil {
+			slog.Warn("could not cleanup stale runners", "path", runners, "error", err)
+		}
+
+		if err := os.Remove(filepath.Dir(match)); err != nil {
+			slog.Warn("could not cleanup stale tmpdir", "path", filepath.Dir(match), "error", err)
 		}
 	}
 }

gpu/gpu.go

@@ -305,38 +305,41 @@ func GetGPUInfo() GpuInfoList {
 		// Intel
 		if envconfig.IntelGPU() {
 			oHandles = initOneAPIHandles()
-			// On windows we bundle the oneapi library one level above the runner dir
-			depPath = ""
-			if runtime.GOOS == "windows" && envconfig.RunnersDir() != "" {
-				depPath = filepath.Join(filepath.Dir(envconfig.RunnersDir()), "oneapi")
-			}
+			if oHandles != nil && oHandles.oneapi != nil {
 
-			for d := range oHandles.oneapi.num_drivers {
-				if oHandles.oneapi == nil {
-					// shouldn't happen
-					slog.Warn("nil oneapi handle with driver count", "count", int(oHandles.oneapi.num_drivers))
-					continue
+				// On windows we bundle the oneapi library one level above the runner dir
+				depPath = ""
+				if runtime.GOOS == "windows" && envconfig.RunnersDir() != "" {
+					depPath = filepath.Join(filepath.Dir(envconfig.RunnersDir()), "oneapi")
 				}
-				devCount := C.oneapi_get_device_count(*oHandles.oneapi, C.int(d))
-				for i := range devCount {
-					gpuInfo := OneapiGPUInfo{
-						GpuInfo: GpuInfo{
-							Library: "oneapi",
-						},
-						driverIndex: int(d),
-						gpuIndex:    int(i),
+
+				for d := range oHandles.oneapi.num_drivers {
+					if oHandles.oneapi == nil {
+						// shouldn't happen
+						slog.Warn("nil oneapi handle with driver count", "count", int(oHandles.oneapi.num_drivers))
+						continue
+					}
+					devCount := C.oneapi_get_device_count(*oHandles.oneapi, C.int(d))
+					for i := range devCount {
+						gpuInfo := OneapiGPUInfo{
+							GpuInfo: GpuInfo{
+								Library: "oneapi",
+							},
+							driverIndex: int(d),
+							gpuIndex:    int(i),
+						}
+						// TODO - split bootstrapping from updating free memory
+						C.oneapi_check_vram(*oHandles.oneapi, C.int(d), i, &memInfo)
+						// TODO - convert this to MinimumMemory based on testing...
+						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 = C.GoString(&memInfo.gpu_id[0])
+						gpuInfo.Name = C.GoString(&memInfo.gpu_name[0])
+						gpuInfo.DependencyPath = depPath
+						oneapiGPUs = append(oneapiGPUs, gpuInfo)
 					}
-					// TODO - split bootstrapping from updating free memory
-					C.oneapi_check_vram(*oHandles.oneapi, C.int(d), i, &memInfo)
-					// TODO - convert this to MinimumMemory based on testing...
-					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 = C.GoString(&memInfo.gpu_id[0])
-					gpuInfo.Name = C.GoString(&memInfo.gpu_name[0])
-					gpuInfo.DependencyPath = depPath
-					oneapiGPUs = append(oneapiGPUs, gpuInfo)
 				}
 			}
 		}

llm/ext_server/server.cpp

@@ -1223,9 +1223,7 @@ struct llama_server_context
 
                 res.result_json = json
                 {
-                    {"id", res.id},
                     {"embedding", std::vector<float>(embd, embd + n_embd)},
-                    {"timings",             slot.get_formated_timings()},
                 };
             }
         }
@@ -3194,41 +3192,17 @@ int main(int argc, char **argv) {
                     prompt = "";
                 }
 
-                if (prompt.size() == 1) {
-                    prompt = prompt[0];
-                }
-
                 // create and queue the task
-                json responses;
-                {
-                    const int id_task = llama.queue_tasks.get_new_id();
-                    llama.queue_results.add_waiting_task_id(id_task);
-                    llama.request_completion(id_task, {{"prompt", prompt}}, true, -1);
+                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}}, true, -1);
 
-                    // get the result
-                    task_result result = llama.queue_results.recv(id_task);
-                    llama.queue_results.remove_waiting_task_id(id_task);
-                    if (result.error) {
-                        return res.set_content(result.result_json.dump(), "application/json; charset=utf-8");
-                    }
+                // get the result
+                task_result result = llama.queue_results.recv(task_id);
+                llama.queue_results.remove_waiting_task_id(task_id);
 
-                    responses = result.result_json.value("results", std::vector<json>{result.result_json});
-                    std::sort(responses.begin(), responses.end(), [](const json& a, const json& b) {
-                        return a["id"] < b["id"];
-                    });
-
-                    json embeddings = json::array();
-
-                    int prompt_n = 0;
-                    for (auto & elem : responses) {
-                        embeddings.push_back(elem.at("embedding"));
-                        prompt_n += elem.at("timings").at("prompt_n").get<int>();
-                    }
-
-                    // send the result
-                    json embedding_res = json{{"embedding", embeddings}, {"prompt_n", prompt_n}};
-                    return res.set_content(embedding_res.dump(), "application/json; charset=utf-8");
-                }
+                // send the result
+                return res.set_content(result.result_json.dump(), "application/json; charset=utf-8");
             });
 
     // GG: if I put the main loop inside a thread, it crashes on the first request when build in Debug!?

llm/server.go

@@ -33,7 +33,7 @@ type LlamaServer interface {
 	Ping(ctx context.Context) error
 	WaitUntilRunning(ctx context.Context) error
 	Completion(ctx context.Context, req CompletionRequest, fn func(CompletionResponse)) error
-	Embed(ctx context.Context, input []string) (*EmbedResponse, error)
+	Embedding(ctx context.Context, input string) ([]float32, error)
 	Tokenize(ctx context.Context, content string) ([]int, error)
 	Detokenize(ctx context.Context, tokens []int) (string, error)
 	Close() error
@@ -125,8 +125,9 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
 		}
 	}
 
-	// On linux, over-allocating CPU memory will almost always result in an error
-	if runtime.GOOS == "linux" {
+	// On linux and windows, over-allocating CPU memory will almost always result in an error
+	// Darwin has fully dynamic swap so has no direct concept of free swap space
+	if runtime.GOOS != "darwin" {
 		systemMemoryRequired := estimate.TotalSize - estimate.VRAMSize
 		available := systemFreeMemory + systemSwapFreeMemory
 		if systemMemoryRequired > available {
@@ -882,24 +883,20 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 	return nil
 }
 
-type EmbedRequest struct {
-	Content []string `json:"content"`
+type EmbeddingRequest struct {
+	Content string `json:"content"`
 }
 
-type EmbedResponse struct {
-	Embedding       [][]float32 `json:"embedding"`
-	PromptEvalCount int         `json:"prompt_n"`
+type EmbeddingResponse struct {
+	Embedding []float32 `json:"embedding"`
 }
 
-func (s *llmServer) Embed(ctx context.Context, input []string) (*EmbedResponse, error) {
-	// each input will use a slot, so we need to acquire the semaphore for
-	// the number of inputs up to numParallel
-	slots := int64(min(len(input), s.numParallel))
-	if err := s.sem.Acquire(ctx, slots); err != nil {
+func (s *llmServer) Embedding(ctx context.Context, input string) ([]float32, error) {
+	if err := s.sem.Acquire(ctx, 1); err != nil {
 		slog.Error("Failed to acquire semaphore", "error", err)
 		return nil, err
 	}
-	defer s.sem.Release(slots)
+	defer s.sem.Release(1)
 
 	// Make sure the server is ready
 	status, err := s.getServerStatusRetry(ctx)
@@ -909,18 +906,18 @@ func (s *llmServer) Embed(ctx context.Context, input []string) (*EmbedResponse,
 		return nil, fmt.Errorf("unexpected server status: %s", status.ToString())
 	}
 
-	data, err := json.Marshal(EmbedRequest{Content: input})
+	data, err := json.Marshal(EmbeddingRequest{Content: input})
 	if err != nil {
 		return nil, fmt.Errorf("error marshaling embed data: %w", err)
 	}
 
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, fmt.Sprintf("http://127.0.0.1:%d/embedding", s.port), bytes.NewBuffer(data))
+	r, err := http.NewRequestWithContext(ctx, http.MethodPost, fmt.Sprintf("http://127.0.0.1:%d/embedding", s.port), bytes.NewBuffer(data))
 	if err != nil {
 		return nil, fmt.Errorf("error creating embed request: %w", err)
 	}
-	req.Header.Set("Content-Type", "application/json")
+	r.Header.Set("Content-Type", "application/json")
 
-	resp, err := http.DefaultClient.Do(req)
+	resp, err := http.DefaultClient.Do(r)
 	if err != nil {
 		return nil, fmt.Errorf("do embedding request: %w", err)
 	}
@@ -936,12 +933,12 @@ func (s *llmServer) Embed(ctx context.Context, input []string) (*EmbedResponse,
 		return nil, fmt.Errorf("%s", body)
 	}
 
-	var e EmbedResponse
+	var e EmbeddingResponse
 	if err := json.Unmarshal(body, &e); err != nil {
 		return nil, fmt.Errorf("unmarshal tokenize response: %w", err)
 	}
 
-	return &e, nil
+	return e.Embedding, nil
 }
 
 type TokenizeRequest struct {

llm/status.go

@@ -26,6 +26,7 @@ var errorPrefixes = []string{
 	"cudaMalloc failed",
 	"\"ERR\"",
 	"error loading model",
+	"GGML_ASSERT",
 }
 
 func (w *StatusWriter) Write(b []byte) (int, error) {

openai/openai_test.go

@@ -7,27 +7,22 @@ import (
 	"io"
 	"net/http"
 	"net/http/httptest"
+	"reflect"
 	"strings"
 	"testing"
 	"time"
 
 	"github.com/gin-gonic/gin"
-	"github.com/stretchr/testify/assert"
 
 	"github.com/ollama/ollama/api"
 )
 
 const (
-	prefix   = `data:image/jpeg;base64,`
-	image    = `iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=`
-	imageURL = prefix + image
+	prefix = `data:image/jpeg;base64,`
+	image  = `iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=`
 )
 
-func prepareRequest(req *http.Request, body any) {
-	bodyBytes, _ := json.Marshal(body)
-	req.Body = io.NopCloser(bytes.NewReader(bodyBytes))
-	req.Header.Set("Content-Type", "application/json")
-}
+var False = false
 
 func captureRequestMiddleware(capturedRequest any) gin.HandlerFunc {
 	return func(c *gin.Context) {
@@ -43,134 +38,136 @@ func captureRequestMiddleware(capturedRequest any) gin.HandlerFunc {
 
 func TestChatMiddleware(t *testing.T) {
 	type testCase struct {
-		Name     string
-		Setup    func(t *testing.T, req *http.Request)
-		Expected func(t *testing.T, req *api.ChatRequest, resp *httptest.ResponseRecorder)
+		name string
+		body string
+		req  api.ChatRequest
+		err  ErrorResponse
 	}
 
 	var capturedRequest *api.ChatRequest
 
 	testCases := []testCase{
 		{
-			Name: "chat handler",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := ChatCompletionRequest{
-					Model:    "test-model",
-					Messages: []Message{{Role: "user", Content: "Hello"}},
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.ChatRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != http.StatusOK {
-					t.Fatalf("expected 200, got %d", resp.Code)
-				}
-
-				if req.Messages[0].Role != "user" {
-					t.Fatalf("expected 'user', got %s", req.Messages[0].Role)
-				}
-
-				if req.Messages[0].Content != "Hello" {
-					t.Fatalf("expected 'Hello', got %s", req.Messages[0].Content)
-				}
+			name: "chat handler",
+			body: `{
+				"model": "test-model",
+				"messages": [
+					{"role": "user", "content": "Hello"}
+				]
+			}`,
+			req: api.ChatRequest{
+				Model: "test-model",
+				Messages: []api.Message{
+					{
+						Role:    "user",
+						Content: "Hello",
+					},
+				},
+				Options: map[string]any{
+					"temperature": 1.0,
+					"top_p":       1.0,
+				},
+				Stream: &False,
 			},
 		},
 		{
-			Name: "chat handler with image content",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := ChatCompletionRequest{
-					Model: "test-model",
-					Messages: []Message{
-						{
-							Role: "user", Content: []map[string]any{
-								{"type": "text", "text": "Hello"},
-								{"type": "image_url", "image_url": map[string]string{"url": imageURL}},
+			name: "chat handler with image content",
+			body: `{
+				"model": "test-model",
+				"messages": [
+					{
+						"role": "user",
+						"content": [
+							{
+								"type": "text",
+								"text": "Hello"
+							},
+							{
+								"type": "image_url",
+								"image_url": {
+									"url": "` + prefix + image + `"
+								}
+							}
+						]
+					}
+				]
+			}`,
+			req: api.ChatRequest{
+				Model: "test-model",
+				Messages: []api.Message{
+					{
+						Role:    "user",
+						Content: "Hello",
+					},
+					{
+						Role: "user",
+						Images: []api.ImageData{
+							func() []byte {
+								img, _ := base64.StdEncoding.DecodeString(image)
+								return img
+							}(),
+						},
+					},
+				},
+				Options: map[string]any{
+					"temperature": 1.0,
+					"top_p":       1.0,
+				},
+				Stream: &False,
+			},
+		},
+		{
+			name: "chat handler with tools",
+			body: `{
+				"model": "test-model",
+				"messages": [
+					{"role": "user", "content": "What's the weather like in Paris Today?"},
+					{"role": "assistant", "tool_calls": [{"id": "id", "type": "function", "function": {"name": "get_current_weather", "arguments": "{\"location\": \"Paris, France\", \"format\": \"celsius\"}"}}]}
+				]
+			}`,
+			req: api.ChatRequest{
+				Model: "test-model",
+				Messages: []api.Message{
+					{
+						Role:    "user",
+						Content: "What's the weather like in Paris Today?",
+					},
+					{
+						Role: "assistant",
+						ToolCalls: []api.ToolCall{
+							{
+								Function: api.ToolCallFunction{
+									Name: "get_current_weather",
+									Arguments: map[string]interface{}{
+										"location": "Paris, France",
+										"format":   "celsius",
+									},
+								},
 							},
 						},
 					},
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.ChatRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != http.StatusOK {
-					t.Fatalf("expected 200, got %d", resp.Code)
-				}
-
-				if req.Messages[0].Role != "user" {
-					t.Fatalf("expected 'user', got %s", req.Messages[0].Role)
-				}
-
-				if req.Messages[0].Content != "Hello" {
-					t.Fatalf("expected 'Hello', got %s", req.Messages[0].Content)
-				}
-
-				img, _ := base64.StdEncoding.DecodeString(imageURL[len(prefix):])
-
-				if req.Messages[1].Role != "user" {
-					t.Fatalf("expected 'user', got %s", req.Messages[1].Role)
-				}
-
-				if !bytes.Equal(req.Messages[1].Images[0], img) {
-					t.Fatalf("expected image encoding, got %s", req.Messages[1].Images[0])
-				}
+				},
+				Options: map[string]any{
+					"temperature": 1.0,
+					"top_p":       1.0,
+				},
+				Stream: &False,
 			},
 		},
+
 		{
-			Name: "chat handler with tools",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := ChatCompletionRequest{
-					Model: "test-model",
-					Messages: []Message{
-						{Role: "user", Content: "What's the weather like in Paris Today?"},
-						{Role: "assistant", ToolCalls: []ToolCall{{
-							ID:   "id",
-							Type: "function",
-							Function: struct {
-								Name      string `json:"name"`
-								Arguments string `json:"arguments"`
-							}{
-								Name:      "get_current_weather",
-								Arguments: "{\"location\": \"Paris, France\", \"format\": \"celsius\"}",
-							},
-						}}},
-					},
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.ChatRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != 200 {
-					t.Fatalf("expected 200, got %d", resp.Code)
-				}
-
-				if req.Messages[0].Content != "What's the weather like in Paris Today?" {
-					t.Fatalf("expected What's the weather like in Paris Today?, got %s", req.Messages[0].Content)
-				}
-
-				if req.Messages[1].ToolCalls[0].Function.Arguments["location"] != "Paris, France" {
-					t.Fatalf("expected 'Paris, France', got %v", req.Messages[1].ToolCalls[0].Function.Arguments["location"])
-				}
-
-				if req.Messages[1].ToolCalls[0].Function.Arguments["format"] != "celsius" {
-					t.Fatalf("expected celsius, got %v", req.Messages[1].ToolCalls[0].Function.Arguments["format"])
-				}
-			},
-		},
-		{
-			Name: "chat handler error forwarding",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := ChatCompletionRequest{
-					Model:    "test-model",
-					Messages: []Message{{Role: "user", Content: 2}},
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.ChatRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != http.StatusBadRequest {
-					t.Fatalf("expected 400, got %d", resp.Code)
-				}
-
-				if !strings.Contains(resp.Body.String(), "invalid message content type") {
-					t.Fatalf("error was not forwarded")
-				}
+			name: "chat handler error forwarding",
+			body: `{
+				"model": "test-model",
+				"messages": [
+					{"role": "user", "content": 2}
+				]
+			}`,
+			err: ErrorResponse{
+				Error: Error{
+					Message: "invalid message content type: float64",
+					Type:    "invalid_request_error",
+				},
 			},
 		},
 	}
@@ -185,16 +182,26 @@ func TestChatMiddleware(t *testing.T) {
 	router.Handle(http.MethodPost, "/api/chat", endpoint)
 
 	for _, tc := range testCases {
-		t.Run(tc.Name, func(t *testing.T) {
-			req, _ := http.NewRequest(http.MethodPost, "/api/chat", nil)
-
-			tc.Setup(t, req)
+		t.Run(tc.name, func(t *testing.T) {
+			req, _ := http.NewRequest(http.MethodPost, "/api/chat", strings.NewReader(tc.body))
+			req.Header.Set("Content-Type", "application/json")
 
 			resp := httptest.NewRecorder()
 			router.ServeHTTP(resp, req)
 
-			tc.Expected(t, capturedRequest, resp)
+			var errResp ErrorResponse
+			if resp.Code != http.StatusOK {
+				if err := json.Unmarshal(resp.Body.Bytes(), &errResp); err != nil {
+					t.Fatal(err)
+				}
+			}
+			if capturedRequest != nil && !reflect.DeepEqual(tc.req, *capturedRequest) {
+				t.Fatal("requests did not match")
+			}
 
+			if !reflect.DeepEqual(tc.err, errResp) {
+				t.Fatal("errors did not match")
+			}
 			capturedRequest = nil
 		})
 	}
@@ -202,71 +209,52 @@ func TestChatMiddleware(t *testing.T) {
 
 func TestCompletionsMiddleware(t *testing.T) {
 	type testCase struct {
-		Name     string
-		Setup    func(t *testing.T, req *http.Request)
-		Expected func(t *testing.T, req *api.GenerateRequest, resp *httptest.ResponseRecorder)
+		name string
+		body string
+		req  api.GenerateRequest
+		err  ErrorResponse
 	}
 
 	var capturedRequest *api.GenerateRequest
 
 	testCases := []testCase{
 		{
-			Name: "completions handler",
-			Setup: func(t *testing.T, req *http.Request) {
-				temp := float32(0.8)
-				body := CompletionRequest{
-					Model:       "test-model",
-					Prompt:      "Hello",
-					Temperature: &temp,
-					Stop:        []string{"\n", "stop"},
-					Suffix:      "suffix",
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.GenerateRequest, resp *httptest.ResponseRecorder) {
-				if req.Prompt != "Hello" {
-					t.Fatalf("expected 'Hello', got %s", req.Prompt)
-				}
-
-				if req.Options["temperature"] != 1.6 {
-					t.Fatalf("expected 1.6, got %f", req.Options["temperature"])
-				}
-
-				stopTokens, ok := req.Options["stop"].([]any)
-
-				if !ok {
-					t.Fatalf("expected stop tokens to be a list")
-				}
-
-				if stopTokens[0] != "\n" || stopTokens[1] != "stop" {
-					t.Fatalf("expected ['\\n', 'stop'], got %v", stopTokens)
-				}
-
-				if req.Suffix != "suffix" {
-					t.Fatalf("expected 'suffix', got %s", req.Suffix)
-				}
+			name: "completions handler",
+			body: `{
+				"model": "test-model",
+				"prompt": "Hello",
+				"temperature": 0.8,
+				"stop": ["\n", "stop"],
+				"suffix": "suffix"
+			}`,
+			req: api.GenerateRequest{
+				Model:  "test-model",
+				Prompt: "Hello",
+				Options: map[string]any{
+					"frequency_penalty": 0.0,
+					"presence_penalty":  0.0,
+					"temperature":       1.6,
+					"top_p":             1.0,
+					"stop":              []any{"\n", "stop"},
+				},
+				Suffix: "suffix",
+				Stream: &False,
 			},
 		},
 		{
-			Name: "completions handler error forwarding",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := CompletionRequest{
-					Model:       "test-model",
-					Prompt:      "Hello",
-					Temperature: nil,
-					Stop:        []int{1, 2},
-					Suffix:      "suffix",
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.GenerateRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != http.StatusBadRequest {
-					t.Fatalf("expected 400, got %d", resp.Code)
-				}
-
-				if !strings.Contains(resp.Body.String(), "invalid type for 'stop' field") {
-					t.Fatalf("error was not forwarded")
-				}
+			name: "completions handler error forwarding",
+			body: `{
+				"model": "test-model",
+				"prompt": "Hello",
+				"temperature": null,
+				"stop": [1, 2],
+				"suffix": "suffix"
+			}`,
+			err: ErrorResponse{
+				Error: Error{
+					Message: "invalid type for 'stop' field: float64",
+					Type:    "invalid_request_error",
+				},
 			},
 		},
 	}
@@ -281,15 +269,27 @@ func TestCompletionsMiddleware(t *testing.T) {
 	router.Handle(http.MethodPost, "/api/generate", endpoint)
 
 	for _, tc := range testCases {
-		t.Run(tc.Name, func(t *testing.T) {
-			req, _ := http.NewRequest(http.MethodPost, "/api/generate", nil)
-
-			tc.Setup(t, req)
+		t.Run(tc.name, func(t *testing.T) {
+			req, _ := http.NewRequest(http.MethodPost, "/api/generate", strings.NewReader(tc.body))
+			req.Header.Set("Content-Type", "application/json")
 
 			resp := httptest.NewRecorder()
 			router.ServeHTTP(resp, req)
 
-			tc.Expected(t, capturedRequest, resp)
+			var errResp ErrorResponse
+			if resp.Code != http.StatusOK {
+				if err := json.Unmarshal(resp.Body.Bytes(), &errResp); err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			if capturedRequest != nil && !reflect.DeepEqual(tc.req, *capturedRequest) {
+				t.Fatal("requests did not match")
+			}
+
+			if !reflect.DeepEqual(tc.err, errResp) {
+				t.Fatal("errors did not match")
+			}
 
 			capturedRequest = nil
 		})
@@ -298,78 +298,47 @@ func TestCompletionsMiddleware(t *testing.T) {
 
 func TestEmbeddingsMiddleware(t *testing.T) {
 	type testCase struct {
-		Name     string
-		Setup    func(t *testing.T, req *http.Request)
-		Expected func(t *testing.T, req *api.EmbedRequest, resp *httptest.ResponseRecorder)
+		name string
+		body string
+		req  api.EmbedRequest
+		err  ErrorResponse
 	}
 
 	var capturedRequest *api.EmbedRequest
 
 	testCases := []testCase{
 		{
-			Name: "embed handler single input",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := EmbedRequest{
-					Input: "Hello",
-					Model: "test-model",
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.EmbedRequest, resp *httptest.ResponseRecorder) {
-				if req.Input != "Hello" {
-					t.Fatalf("expected 'Hello', got %s", req.Input)
-				}
-
-				if req.Model != "test-model" {
-					t.Fatalf("expected 'test-model', got %s", req.Model)
-				}
+			name: "embed handler single input",
+			body: `{
+				"input": "Hello",
+				"model": "test-model"
+			}`,
+			req: api.EmbedRequest{
+				Input: "Hello",
+				Model: "test-model",
 			},
 		},
 		{
-			Name: "embed handler batch input",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := EmbedRequest{
-					Input: []string{"Hello", "World"},
-					Model: "test-model",
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.EmbedRequest, resp *httptest.ResponseRecorder) {
-				input, ok := req.Input.([]any)
-
-				if !ok {
-					t.Fatalf("expected input to be a list")
-				}
-
-				if input[0].(string) != "Hello" {
-					t.Fatalf("expected 'Hello', got %s", input[0])
-				}
-
-				if input[1].(string) != "World" {
-					t.Fatalf("expected 'World', got %s", input[1])
-				}
-
-				if req.Model != "test-model" {
-					t.Fatalf("expected 'test-model', got %s", req.Model)
-				}
+			name: "embed handler batch input",
+			body: `{
+				"input": ["Hello", "World"],
+				"model": "test-model"
+			}`,
+			req: api.EmbedRequest{
+				Input: []any{"Hello", "World"},
+				Model: "test-model",
 			},
 		},
 		{
-			Name: "embed handler error forwarding",
-			Setup: func(t *testing.T, req *http.Request) {
-				body := EmbedRequest{
-					Model: "test-model",
-				}
-				prepareRequest(req, body)
-			},
-			Expected: func(t *testing.T, req *api.EmbedRequest, resp *httptest.ResponseRecorder) {
-				if resp.Code != http.StatusBadRequest {
-					t.Fatalf("expected 400, got %d", resp.Code)
-				}
-
-				if !strings.Contains(resp.Body.String(), "invalid input") {
-					t.Fatalf("error was not forwarded")
-				}
+			name: "embed handler error forwarding",
+			body: `{
+				"model": "test-model"
+			}`,
+			err: ErrorResponse{
+				Error: Error{
+					Message: "invalid input",
+					Type:    "invalid_request_error",
+				},
 			},
 		},
 	}
@@ -384,116 +353,167 @@ func TestEmbeddingsMiddleware(t *testing.T) {
 	router.Handle(http.MethodPost, "/api/embed", endpoint)
 
 	for _, tc := range testCases {
-		t.Run(tc.Name, func(t *testing.T) {
-			req, _ := http.NewRequest(http.MethodPost, "/api/embed", nil)
-
-			tc.Setup(t, req)
+		t.Run(tc.name, func(t *testing.T) {
+			req, _ := http.NewRequest(http.MethodPost, "/api/embed", strings.NewReader(tc.body))
+			req.Header.Set("Content-Type", "application/json")
 
 			resp := httptest.NewRecorder()
 			router.ServeHTTP(resp, req)
 
-			tc.Expected(t, capturedRequest, resp)
+			var errResp ErrorResponse
+			if resp.Code != http.StatusOK {
+				if err := json.Unmarshal(resp.Body.Bytes(), &errResp); err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			if capturedRequest != nil && !reflect.DeepEqual(tc.req, *capturedRequest) {
+				t.Fatal("requests did not match")
+			}
+
+			if !reflect.DeepEqual(tc.err, errResp) {
+				t.Fatal("errors did not match")
+			}
 
 			capturedRequest = nil
 		})
 	}
 }
 
-func TestMiddlewareResponses(t *testing.T) {
+func TestListMiddleware(t *testing.T) {
 	type testCase struct {
-		Name     string
-		Method   string
-		Path     string
-		TestPath string
-		Handler  func() gin.HandlerFunc
-		Endpoint func(c *gin.Context)
-		Setup    func(t *testing.T, req *http.Request)
-		Expected func(t *testing.T, resp *httptest.ResponseRecorder)
+		name     string
+		endpoint func(c *gin.Context)
+		resp     string
 	}
 
 	testCases := []testCase{
 		{
-			Name:     "list handler",
-			Method:   http.MethodGet,
-			Path:     "/api/tags",
-			TestPath: "/api/tags",
-			Handler:  ListMiddleware,
-			Endpoint: func(c *gin.Context) {
+			name: "list handler",
+			endpoint: func(c *gin.Context) {
 				c.JSON(http.StatusOK, api.ListResponse{
 					Models: []api.ListModelResponse{
 						{
-							Name: "Test Model",
+							Name:       "test-model",
+							ModifiedAt: time.Unix(int64(1686935002), 0).UTC(),
 						},
 					},
 				})
 			},
-			Expected: func(t *testing.T, resp *httptest.ResponseRecorder) {
-				var listResp ListCompletion
-				if err := json.NewDecoder(resp.Body).Decode(&listResp); err != nil {
-					t.Fatal(err)
-				}
-
-				if listResp.Object != "list" {
-					t.Fatalf("expected list, got %s", listResp.Object)
-				}
-
-				if len(listResp.Data) != 1 {
-					t.Fatalf("expected 1, got %d", len(listResp.Data))
-				}
-
-				if listResp.Data[0].Id != "Test Model" {
-					t.Fatalf("expected Test Model, got %s", listResp.Data[0].Id)
-				}
-			},
+			resp: `{
+				"object": "list",
+				"data": [
+					{
+						"id": "test-model",
+						"object": "model",
+						"created": 1686935002,
+						"owned_by": "library"
+					}
+				]
+			}`,
 		},
 		{
-			Name:     "retrieve model",
-			Method:   http.MethodGet,
-			Path:     "/api/show/:model",
-			TestPath: "/api/show/test-model",
-			Handler:  RetrieveMiddleware,
-			Endpoint: func(c *gin.Context) {
-				c.JSON(http.StatusOK, api.ShowResponse{
-					ModifiedAt: time.Date(2024, 6, 17, 13, 45, 0, 0, time.UTC),
-				})
-			},
-			Expected: func(t *testing.T, resp *httptest.ResponseRecorder) {
-				var retrieveResp Model
-				if err := json.NewDecoder(resp.Body).Decode(&retrieveResp); err != nil {
-					t.Fatal(err)
-				}
-
-				if retrieveResp.Object != "model" {
-					t.Fatalf("Expected object to be model, got %s", retrieveResp.Object)
-				}
-
-				if retrieveResp.Id != "test-model" {
-					t.Fatalf("Expected id to be test-model, got %s", retrieveResp.Id)
-				}
+			name: "list handler empty output",
+			endpoint: func(c *gin.Context) {
+				c.JSON(http.StatusOK, api.ListResponse{})
 			},
+			resp: `{
+				"object": "list",
+				"data": null
+			}`,
 		},
 	}
 
 	gin.SetMode(gin.TestMode)
-	router := gin.New()
 
 	for _, tc := range testCases {
-		t.Run(tc.Name, func(t *testing.T) {
-			router = gin.New()
-			router.Use(tc.Handler())
-			router.Handle(tc.Method, tc.Path, tc.Endpoint)
-			req, _ := http.NewRequest(tc.Method, tc.TestPath, nil)
+		router := gin.New()
+		router.Use(ListMiddleware())
+		router.Handle(http.MethodGet, "/api/tags", tc.endpoint)
+		req, _ := http.NewRequest(http.MethodGet, "/api/tags", nil)
 
-			if tc.Setup != nil {
-				tc.Setup(t, req)
-			}
+		resp := httptest.NewRecorder()
+		router.ServeHTTP(resp, req)
 
-			resp := httptest.NewRecorder()
-			router.ServeHTTP(resp, req)
+		var expected, actual map[string]any
+		err := json.Unmarshal([]byte(tc.resp), &expected)
+		if err != nil {
+			t.Fatalf("failed to unmarshal expected response: %v", err)
+		}
 
-			assert.Equal(t, http.StatusOK, resp.Code)
+		err = json.Unmarshal(resp.Body.Bytes(), &actual)
+		if err != nil {
+			t.Fatalf("failed to unmarshal actual response: %v", err)
+		}
 
-			tc.Expected(t, resp)
-		})
+		if !reflect.DeepEqual(expected, actual) {
+			t.Errorf("responses did not match\nExpected: %+v\nActual: %+v", expected, actual)
+		}
+	}
+}
+
+func TestRetrieveMiddleware(t *testing.T) {
+	type testCase struct {
+		name     string
+		endpoint func(c *gin.Context)
+		resp     string
+	}
+
+	testCases := []testCase{
+		{
+			name: "retrieve handler",
+			endpoint: func(c *gin.Context) {
+				c.JSON(http.StatusOK, api.ShowResponse{
+					ModifiedAt: time.Unix(int64(1686935002), 0).UTC(),
+				})
+			},
+			resp: `{
+				"id":"test-model",
+				"object":"model",
+				"created":1686935002,
+				"owned_by":"library"}
+			`,
+		},
+		{
+			name: "retrieve handler error forwarding",
+			endpoint: func(c *gin.Context) {
+				c.JSON(http.StatusBadRequest, gin.H{"error": "model not found"})
+			},
+			resp: `{
+				"error": {
+				  "code": null,
+				  "message": "model not found",
+				  "param": null,
+				  "type": "api_error"
+				}
+			}`,
+		},
+	}
+
+	gin.SetMode(gin.TestMode)
+
+	for _, tc := range testCases {
+		router := gin.New()
+		router.Use(RetrieveMiddleware())
+		router.Handle(http.MethodGet, "/api/show/:model", tc.endpoint)
+		req, _ := http.NewRequest(http.MethodGet, "/api/show/test-model", nil)
+
+		resp := httptest.NewRecorder()
+		router.ServeHTTP(resp, req)
+
+		var expected, actual map[string]any
+		err := json.Unmarshal([]byte(tc.resp), &expected)
+		if err != nil {
+			t.Fatalf("failed to unmarshal expected response: %v", err)
+		}
+
+		err = json.Unmarshal(resp.Body.Bytes(), &actual)
+		if err != nil {
+			t.Fatalf("failed to unmarshal actual response: %v", err)
+		}
+
+		if !reflect.DeepEqual(expected, actual) {
+			t.Errorf("responses did not match\nExpected: %+v\nActual: %+v", expected, actual)
+		}
 	}
 }

scripts/install.sh

@@ -209,15 +209,15 @@ install_cuda_driver_yum() {
     case $PACKAGE_MANAGER in
         yum)
             $SUDO $PACKAGE_MANAGER -y install yum-utils
-            if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-$1$2.repo" >/dev/null ; then
-                $SUDO $PACKAGE_MANAGER-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-$1$2.repo
+            if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-$1$2.repo" >/dev/null ; then
+                $SUDO $PACKAGE_MANAGER-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-$1$2.repo
             else
                 error $CUDA_REPO_ERR_MSG
             fi
             ;;
         dnf)
-            if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-$1$2.repo" >/dev/null ; then
-                $SUDO $PACKAGE_MANAGER config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-$1$2.repo
+            if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-$1$2.repo" >/dev/null ; then
+                $SUDO $PACKAGE_MANAGER config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-$1$2.repo
             else
                 error $CUDA_REPO_ERR_MSG
             fi
@@ -245,8 +245,8 @@ install_cuda_driver_yum() {
 # ref: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#debian
 install_cuda_driver_apt() {
     status 'Installing NVIDIA repository...'
-    if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-keyring_1.1-1_all.deb" >/dev/null ; then
-        curl -fsSL -o $TEMP_DIR/cuda-keyring.deb https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m)/cuda-keyring_1.1-1_all.deb
+    if curl -I --silent --fail --location "https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-keyring_1.1-1_all.deb" >/dev/null ; then
+        curl -fsSL -o $TEMP_DIR/cuda-keyring.deb https://developer.download.nvidia.com/compute/cuda/repos/$1$2/$(uname -m | sed -e 's/aarch64/sbsa/')/cuda-keyring_1.1-1_all.deb
     else
         error $CUDA_REPO_ERR_MSG
     fi

server/download.go

@@ -216,9 +216,7 @@ func (b *blobDownload) run(ctx context.Context, requestURL *url.URL, opts *regis
 		return err
 	}
 	defer file.Close()
-	if err := setSparse(file); err != nil {
-		return err
-	}
+	setSparse(file)
 
 	_ = file.Truncate(b.Total)
 
@@ -235,7 +233,7 @@ func (b *blobDownload) run(ctx context.Context, requestURL *url.URL, opts *regis
 
 			newOpts.CheckRedirect = func(req *http.Request, via []*http.Request) error {
 				if len(via) > 10 {
-					return errors.New("maxium redirects exceeded (10) for directURL")
+					return errors.New("maximum redirects exceeded (10) for directURL")
 				}
 
 				// if the hostname is the same, allow the redirect

server/images.go

@@ -250,19 +250,21 @@ func GetModel(name string) (*Model, error) {
 		Template:  template.DefaultTemplate,
 	}
 
-	filename, err := GetBlobsPath(manifest.Config.Digest)
-	if err != nil {
-		return nil, err
-	}
+	if manifest.Config.Digest != "" {
+		filename, err := GetBlobsPath(manifest.Config.Digest)
+		if err != nil {
+			return nil, err
+		}
 
-	configFile, err := os.Open(filename)
-	if err != nil {
-		return nil, err
-	}
-	defer configFile.Close()
+		configFile, err := os.Open(filename)
+		if err != nil {
+			return nil, err
+		}
+		defer configFile.Close()
 
-	if err := json.NewDecoder(configFile).Decode(&model.Config); err != nil {
-		return nil, err
+		if err := json.NewDecoder(configFile).Decode(&model.Config); err != nil {
+			return nil, err
+		}
 	}
 
 	for _, layer := range manifest.Layers {
@@ -371,7 +373,7 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 	var messages []*api.Message
 	parameters := make(map[string]any)
 
-	var layers []*Layer
+	var layers []Layer
 	for _, c := range modelfile.Commands {
 		mediatype := fmt.Sprintf("application/vnd.ollama.image.%s", c.Name)
 
@@ -497,7 +499,7 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 
 			if c.Name != "license" {
 				// replace
-				layers = slices.DeleteFunc(layers, func(layer *Layer) bool {
+				layers = slices.DeleteFunc(layers, func(layer Layer) bool {
 					if layer.MediaType != mediatype {
 						return false
 					}
@@ -543,7 +545,7 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 	}
 
 	var err2 error
-	layers = slices.DeleteFunc(layers, func(layer *Layer) bool {
+	layers = slices.DeleteFunc(layers, func(layer Layer) bool {
 		switch layer.MediaType {
 		case "application/vnd.ollama.image.message":
 			// if there are new messages, remove the inherited ones
@@ -623,12 +625,12 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 		return err
 	}
 
-	layer, err := NewLayer(&b, "application/vnd.docker.container.image.v1+json")
+	configLayer, err := NewLayer(&b, "application/vnd.docker.container.image.v1+json")
 	if err != nil {
 		return err
 	}
 
-	for _, layer := range append(layers, layer) {
+	for _, layer := range append(layers, configLayer) {
 		if layer.status != "" {
 			fn(api.ProgressResponse{Status: layer.status})
 		}
@@ -637,7 +639,7 @@ func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantizatio
 	old, _ := ParseNamedManifest(name)
 
 	fn(api.ProgressResponse{Status: "writing manifest"})
-	if err := WriteManifest(name, layer, layers); err != nil {
+	if err := WriteManifest(name, configLayer, layers); err != nil {
 		return err
 	}
 
@@ -714,8 +716,7 @@ 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
-			return nil
+			return err
 		}
 
 		for _, layer := range manifest.Layers {
@@ -782,7 +783,8 @@ func PruneLayers() error {
 
 	err = deleteUnusedLayers(nil, deleteMap)
 	if err != nil {
-		return err
+		slog.Error(fmt.Sprintf("couldn't remove unused layers: %v", err))
+		return nil
 	}
 
 	slog.Info(fmt.Sprintf("total unused blobs removed: %d", len(deleteMap)))
@@ -837,9 +839,11 @@ func PushModel(ctx context.Context, name string, regOpts *registryOptions, fn fu
 		return err
 	}
 
-	var layers []*Layer
+	var layers []Layer
 	layers = append(layers, manifest.Layers...)
-	layers = append(layers, manifest.Config)
+	if manifest.Config.Digest != "" {
+		layers = append(layers, manifest.Config)
+	}
 
 	for _, layer := range layers {
 		if err := uploadBlob(ctx, mp, layer, regOpts, fn); err != nil {
@@ -890,7 +894,9 @@ func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn fu
 			for _, l := range manifest.Layers {
 				deleteMap[l.Digest] = struct{}{}
 			}
-			deleteMap[manifest.Config.Digest] = struct{}{}
+			if manifest.Config.Digest != "" {
+				deleteMap[manifest.Config.Digest] = struct{}{}
+			}
 		}
 	}
 
@@ -905,9 +911,11 @@ func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn fu
 		return fmt.Errorf("pull model manifest: %s", err)
 	}
 
-	var layers []*Layer
+	var layers []Layer
 	layers = append(layers, manifest.Layers...)
-	layers = append(layers, manifest.Config)
+	if manifest.Config.Digest != "" {
+		layers = append(layers, manifest.Config)
+	}
 
 	skipVerify := make(map[string]bool)
 	for _, layer := range layers {
@@ -971,7 +979,8 @@ func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn fu
 		fn(api.ProgressResponse{Status: "removing any unused layers"})
 		err = deleteUnusedLayers(nil, deleteMap)
 		if err != nil {
-			return err
+			slog.Error(fmt.Sprintf("couldn't remove unused layers: %v", err))
+			fn(api.ProgressResponse{Status: fmt.Sprintf("couldn't remove unused layers: %v", err)})
 		}
 	}
 

server/layer.go

@@ -2,6 +2,7 @@ package server
 
 import (
 	"crypto/sha256"
+	"errors"
 	"fmt"
 	"io"
 	"os"
@@ -15,15 +16,15 @@ type Layer struct {
 	status    string
 }
 
-func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
+func NewLayer(r io.Reader, mediatype string) (Layer, error) {
 	blobs, err := GetBlobsPath("")
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
 	temp, err := os.CreateTemp(blobs, "sha256-")
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 	defer temp.Close()
 	defer os.Remove(temp.Name())
@@ -31,28 +32,28 @@ func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
 	sha256sum := sha256.New()
 	n, err := io.Copy(io.MultiWriter(temp, sha256sum), r)
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
 	if err := temp.Close(); err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
 	digest := fmt.Sprintf("sha256:%x", sha256sum.Sum(nil))
 	blob, err := GetBlobsPath(digest)
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
 	status := "using existing layer"
 	if _, err := os.Stat(blob); err != nil {
 		status = "creating new layer"
 		if err := os.Rename(temp.Name(), blob); err != nil {
-			return nil, err
+			return Layer{}, err
 		}
 	}
 
-	return &Layer{
+	return Layer{
 		MediaType: mediatype,
 		Digest:    digest,
 		Size:      n,
@@ -60,18 +61,22 @@ func NewLayer(r io.Reader, mediatype string) (*Layer, error) {
 	}, nil
 }
 
-func NewLayerFromLayer(digest, mediatype, from string) (*Layer, error) {
+func NewLayerFromLayer(digest, mediatype, from string) (Layer, error) {
+	if digest == "" {
+		return Layer{}, errors.New("creating new layer from layer with empty digest")
+	}
+
 	blob, err := GetBlobsPath(digest)
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
 	fi, err := os.Stat(blob)
 	if err != nil {
-		return nil, err
+		return Layer{}, err
 	}
 
-	return &Layer{
+	return Layer{
 		MediaType: mediatype,
 		Digest:    digest,
 		Size:      fi.Size(),
@@ -81,6 +86,10 @@ func NewLayerFromLayer(digest, mediatype, from string) (*Layer, error) {
 }
 
 func (l *Layer) Open() (io.ReadSeekCloser, error) {
+	if l.Digest == "" {
+		return nil, errors.New("opening layer with empty digest")
+	}
+
 	blob, err := GetBlobsPath(l.Digest)
 	if err != nil {
 		return nil, err
@@ -90,6 +99,10 @@ func (l *Layer) Open() (io.ReadSeekCloser, error) {
 }
 
 func (l *Layer) Remove() error {
+	if l.Digest == "" {
+		return nil
+	}
+
 	ms, err := Manifests()
 	if err != nil {
 		return err

server/manifest.go

@@ -14,10 +14,10 @@ import (
 )
 
 type Manifest struct {
-	SchemaVersion int      `json:"schemaVersion"`
-	MediaType     string   `json:"mediaType"`
-	Config        *Layer   `json:"config"`
-	Layers        []*Layer `json:"layers"`
+	SchemaVersion int     `json:"schemaVersion"`
+	MediaType     string  `json:"mediaType"`
+	Config        Layer   `json:"config"`
+	Layers        []Layer `json:"layers"`
 
 	filepath string
 	fi       os.FileInfo
@@ -47,10 +47,12 @@ func (m *Manifest) Remove() error {
 
 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
+		if layer.Digest != "" {
+			if err := layer.Remove(); errors.Is(err, os.ErrNotExist) {
+				slog.Debug("layer does not exist", "digest", layer.Digest)
+			} else if err != nil {
+				return err
+			}
 		}
 	}
 
@@ -93,7 +95,7 @@ func ParseNamedManifest(n model.Name) (*Manifest, error) {
 	return &m, nil
 }
 
-func WriteManifest(name model.Name, config *Layer, layers []*Layer) error {
+func WriteManifest(name model.Name, config Layer, layers []Layer) error {
 	manifests, err := GetManifestPath()
 	if err != nil {
 		return err

server/model.go

@@ -26,7 +26,7 @@ import (
 var intermediateBlobs map[string]string = make(map[string]string)
 
 type layerGGML struct {
-	*Layer
+	Layer
 	*llm.GGML
 }
 

server/routes.go

@@ -23,6 +23,7 @@ import (
 
 	"github.com/gin-contrib/cors"
 	"github.com/gin-gonic/gin"
+	"golang.org/x/sync/errgroup"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/envconfig"
@@ -346,6 +347,7 @@ func (s *Server) EmbedHandler(c *gin.Context) {
 		return
 	}
 
+	var count int
 	for i, s := range input {
 		tokens, err := r.Tokenize(c.Request.Context(), s)
 		if err != nil {
@@ -368,25 +370,36 @@ func (s *Server) EmbedHandler(c *gin.Context) {
 			}
 		}
 
+		count += len(tokens)
+
 		input[i] = s
 	}
-	embeddings, err := r.Embed(c.Request.Context(), input)
-	if err != nil {
-		slog.Error("embedding generation failed", "error", err)
-		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate embedding"})
-		return
+
+	var g errgroup.Group
+	embeddings := make([][]float32, len(input))
+	for i, text := range input {
+		g.Go(func() error {
+			embedding, err := r.Embedding(c.Request.Context(), text)
+			if err != nil {
+				return err
+			}
+			embeddings[i] = normalize(embedding)
+			return nil
+		})
 	}
 
-	for i, e := range embeddings.Embedding {
-		embeddings.Embedding[i] = normalize(e)
+	if err := g.Wait(); err != nil {
+		slog.Error("embedding generation failed", "error", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Errorf("failed to generate embeddings: %v", err)})
+		return
 	}
 
 	resp := api.EmbedResponse{
 		Model:           req.Model,
-		Embeddings:      embeddings.Embedding,
+		Embeddings:      embeddings,
 		TotalDuration:   time.Since(checkpointStart),
 		LoadDuration:    checkpointLoaded.Sub(checkpointStart),
-		PromptEvalCount: embeddings.PromptEvalCount,
+		PromptEvalCount: count,
 	}
 	c.JSON(http.StatusOK, resp)
 }
@@ -430,21 +443,20 @@ func (s *Server) EmbeddingsHandler(c *gin.Context) {
 		return
 	}
 
-	embeddings, err := r.Embed(c.Request.Context(), []string{req.Prompt})
+	embedding, err := r.Embedding(c.Request.Context(), req.Prompt)
 	if err != nil {
 		slog.Info(fmt.Sprintf("embedding generation failed: %v", err))
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate embedding"})
 		return
 	}
 
-	embedding := make([]float64, len(embeddings.Embedding[0]))
-
-	for i, v := range embeddings.Embedding[0] {
-		embedding[i] = float64(v)
+	var e []float64
+	for _, v := range embedding {
+		e = append(e, float64(v))
 	}
 
 	resp := api.EmbeddingResponse{
-		Embedding: embedding,
+		Embedding: e,
 	}
 	c.JSON(http.StatusOK, resp)
 }
@@ -824,17 +836,20 @@ func (s *Server) ListModelsHandler(c *gin.Context) {
 
 	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()
-
 		var cf ConfigV2
-		if err := json.NewDecoder(f).Decode(&cf); err != nil {
-			slog.Warn("bad manifest config", "name", n, "error", err)
-			continue
+
+		if m.Config.Digest != "" {
+			f, err := m.Config.Open()
+			if err != nil {
+				slog.Warn("bad manifest filepath", "name", n, "error", err)
+				continue
+			}
+			defer f.Close()
+
+			if err := json.NewDecoder(f).Decode(&cf); err != nil {
+				slog.Warn("bad manifest config", "name", n, "error", err)
+				continue
+			}
 		}
 
 		// tag should never be masked
@@ -1028,6 +1043,11 @@ func allowedHostsMiddleware(addr net.Addr) gin.HandlerFunc {
 
 		if addr, err := netip.ParseAddr(host); err == nil {
 			if addr.IsLoopback() || addr.IsPrivate() || addr.IsUnspecified() || isLocalIP(addr) {
+				if c.Request.Method == http.MethodOptions {
+					c.AbortWithStatus(http.StatusNoContent)
+					return
+				}
+
 				c.Next()
 				return
 			}
@@ -1059,6 +1079,7 @@ func (s *Server) GenerateRoutes() http.Handler {
 	config.AllowOrigins = envconfig.Origins()
 
 	r := gin.Default()
+	r.HandleMethodNotAllowed = true
 	r.Use(
 		cors.New(config),
 		allowedHostsMiddleware(s.addr),

server/routes_delete_test.go

@@ -98,7 +98,7 @@ func TestDeleteDuplicateLayers(t *testing.T) {
 	}
 
 	// create a manifest with duplicate layers
-	if err := WriteManifest(n, config, []*Layer{config}); err != nil {
+	if err := WriteManifest(n, config, []Layer{config}); err != nil {
 		t.Fatal(err)
 	}
 

server/sched_test.go

@@ -708,8 +708,8 @@ type mockLlm struct {
 	pingResp           error
 	waitResp           error
 	completionResp     error
-	embedResp          *llm.EmbedResponse
-	embedRespErr       error
+	embeddingResp      []float32
+	embeddingRespErr   error
 	tokenizeResp       []int
 	tokenizeRespErr    error
 	detokenizeResp     string
@@ -727,8 +727,8 @@ func (s *mockLlm) Completion(ctx context.Context, req llm.CompletionRequest, fn
 	return s.completionResp
 }
 
-func (s *mockLlm) Embed(ctx context.Context, input []string) (*llm.EmbedResponse, error) {
-	return s.embedResp, s.embedRespErr
+func (s *mockLlm) Embedding(ctx context.Context, input string) ([]float32, error) {
+	return s.embeddingResp, s.embeddingRespErr
 }
 
 func (s *mockLlm) Tokenize(ctx context.Context, content string) ([]int, error) {

server/sparse_common.go

@@ -4,6 +4,5 @@ package server
 
 import "os"
 
-func setSparse(file *os.File) error {
-	return nil
+func setSparse(*os.File) {
 }

server/sparse_windows.go

@@ -6,8 +6,9 @@ import (
 	"golang.org/x/sys/windows"
 )
 
-func setSparse(file *os.File) error {
-	return windows.DeviceIoControl(
+func setSparse(file *os.File) {
+	// exFat (and other FS types) don't support sparse files, so ignore errors
+	windows.DeviceIoControl( //nolint:errcheck
 		windows.Handle(file.Fd()), windows.FSCTL_SET_SPARSE,
 		nil, 0,
 		nil, 0,

server/upload.go

@@ -26,7 +26,7 @@ import (
 var blobUploadManager sync.Map
 
 type blobUpload struct {
-	*Layer
+	Layer
 
 	Total     int64
 	Completed atomic.Int64
@@ -362,7 +362,7 @@ func (p *progressWriter) Rollback() {
 	p.written = 0
 }
 
-func uploadBlob(ctx context.Context, mp ModelPath, layer *Layer, opts *registryOptions, fn func(api.ProgressResponse)) error {
+func uploadBlob(ctx context.Context, mp ModelPath, layer Layer, opts *registryOptions, fn func(api.ProgressResponse)) error {
 	requestURL := mp.BaseURL()
 	requestURL = requestURL.JoinPath("v2", mp.GetNamespaceRepository(), "blobs", layer.Digest)