Initial Draft

api/types.go

@@ -210,7 +210,10 @@ type EmbeddingRequest struct {
 	Model string `json:"model"`
 
 	// Prompt is the textual prompt to embed.
-	Prompt string `json:"prompt"`
+	Prompt string `json:"prompt,omitempty"`
+
+	// PromptBatch is a list of prompts to embed.
+	PromptBatch []string `json:"prompt_batch,omitempty"`
 
 	// KeepAlive controls how long the model will stay loaded in memory following
 	// this request.
@@ -222,7 +225,8 @@ type EmbeddingRequest struct {
 
 // EmbeddingResponse is the response from [Client.Embeddings].
 type EmbeddingResponse struct {
-	Embedding []float64 `json:"embedding"`
+	Embedding      []float64   `json:"embedding,omitempty"`
+	EmbeddingBatch [][]float64 `json:"embedding_batch,omitempty"`
 }
 
 // CreateRequest is the request passed to [Client.Create].

llm/ext_server/server.cpp

@@ -3166,26 +3166,36 @@ int main(int argc, char **argv) {
                     prompt = "";
                 }
 
-                json image_data;
-                if (body.count("image_data") != 0) {
-                    image_data = body["image_data"];
-                }
-                else
+                // create and queue the task
+                json responses;
                 {
-                    image_data = "";
+                    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);
+
+                    // get the result
+                    task_result result = llama.queue_results.recv(id_task);
+                    llama.queue_results.remove_waiting_task_id(id_task);
+                    if (!result.error) {
+                        if (result.result_json.count("results")) {
+                            // result for multi-task
+                            responses = result.result_json.at("results");
+                        } else {
+                            // result for single task
+                            responses = std::vector<json>(1, result.result_json);
+                        }
+                        json embeddings = json::array();
+                        for (auto & elem : responses) {
+                            embeddings.push_back(json_value(elem, "embedding", json::array()));
+                        }
+                        // send the result
+                        json result = json{{"embedding", embeddings}};
+                        return res.set_content(result.dump(), "application/json; charset=utf-8");
+                    } else {
+                        // return error
+                        return res.set_content(result.result_json.dump(), "application/json; charset=utf-8");
+                    }
                 }
-
-                // 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} }, true, -1);
-
-                // get the result
-                task_result result = llama.queue_results.recv(task_id);
-                llama.queue_results.remove_waiting_task_id(task_id);
-
-                // 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
-	Embedding(ctx context.Context, prompt string) ([]float64, error)
+	Embedding(ctx context.Context, prompt []string) ([][]float64, error)
 	Tokenize(ctx context.Context, content string) ([]int, error)
 	Detokenize(ctx context.Context, tokens []int) (string, error)
 	Close() error
@@ -842,14 +842,14 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 }
 
 type EmbeddingRequest struct {
-	Content string `json:"content"`
+	Content []string `json:"content"`
 }
 
 type EmbeddingResponse struct {
-	Embedding []float64 `json:"embedding"`
+	Embedding [][]float64 `json:"embedding"`
 }
 
-func (s *llmServer) Embedding(ctx context.Context, prompt string) ([]float64, error) {
+func (s *llmServer) Embedding(ctx context.Context, prompt []string) ([][]float64, error) {
 	if err := s.sem.Acquire(ctx, 1); err != nil {
 		slog.Error("Failed to acquire semaphore", "error", err)
 		return nil, err
@@ -864,7 +864,7 @@ func (s *llmServer) Embedding(ctx context.Context, prompt string) ([]float64, er
 		return nil, fmt.Errorf("unexpected server status: %s", status.ToString())
 	}
 
-	data, err := json.Marshal(TokenizeRequest{Content: prompt})
+	data, err := json.Marshal(EmbeddingRequest{Content: prompt})
 	if err != nil {
 		return nil, fmt.Errorf("error marshaling embed data: %w", err)
 	}

server/routes.go

@@ -389,23 +389,39 @@ func (s *Server) EmbeddingsHandler(c *gin.Context) {
 		return
 	}
 
-	// an empty request loads the model
-	if req.Prompt == "" {
-		c.JSON(http.StatusOK, api.EmbeddingResponse{Embedding: []float64{}})
-		return
-	}
+	switch {
+	// single embedding
+	case len(req.Prompt) > 0:
+		slog.Info("embedding request", "prompt", req.Prompt)
+		embeddings, err := runner.llama.Embedding(c.Request.Context(), []string{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, err := runner.llama.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
-	}
+		resp := api.EmbeddingResponse{Embedding: embeddings[0]}
+		c.JSON(http.StatusOK, resp)
+	// batch embeddings
+	case len(req.PromptBatch) > 0:
+		embeddings, err := runner.llama.Embedding(c.Request.Context(), req.PromptBatch)
+		if err != nil {
+			slog.Info(fmt.Sprintf("batch embedding generation failed: %v", err))
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate embedding"})
+			return
+		}
+
+		resp := api.EmbeddingResponse{EmbeddingBatch: embeddings}
+		c.JSON(http.StatusOK, resp)
 
-	resp := api.EmbeddingResponse{
-		Embedding: embedding,
+	// empty prompt loads the model
+	default:
+		if req.PromptBatch != nil {
+			c.JSON(http.StatusOK, api.EmbeddingResponse{EmbeddingBatch: [][]float64{}})
+		} else {
+			c.JSON(http.StatusOK, api.EmbeddingResponse{Embedding: []float64{}})
+		}
 	}
-	c.JSON(http.StatusOK, resp)
 }
 
 func (s *Server) PullModelHandler(c *gin.Context) {

server/sched_test.go

@@ -608,7 +608,7 @@ type mockLlm struct {
 	pingResp           error
 	waitResp           error
 	completionResp     error
-	embeddingResp      []float64
+	embeddingResp      [][]float64
 	embeddingRespErr   error
 	tokenizeResp       []int
 	tokenizeRespErr    error
@@ -626,7 +626,7 @@ func (s *mockLlm) WaitUntilRunning(ctx context.Context) error { return s.waitRes
 func (s *mockLlm) Completion(ctx context.Context, req llm.CompletionRequest, fn func(llm.CompletionResponse)) error {
 	return s.completionResp
 }
-func (s *mockLlm) Embedding(ctx context.Context, prompt string) ([]float64, error) {
+func (s *mockLlm) Embedding(ctx context.Context, prompt []string) ([][]float64, error) {
 	return s.embeddingResp, s.embeddingRespErr
 }
 func (s *mockLlm) Tokenize(ctx context.Context, content string) ([]int, error) {