feat: 🎉 Vectors na Vectors, the begining

Translate 1536 <-> 768 , 3072 <-> 2048

pkg/server/openai.go

@@ -0,0 +1,102 @@
+package server
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"time"
+
+	"github.com/uptrace/bunrouter"
+
+	"github.com/Warky-Devs/vecna.git/pkg/embedclient"
+)
+
+type openAIEmbedRequest struct {
+	Input interface{} `json:"input"` // string or []string
+	Model string      `json:"model"`
+}
+
+type openAIEmbedResponse struct {
+	Object string             `json:"object"`
+	Data   []openAIEmbedDatum `json:"data"`
+	Model  string             `json:"model"`
+	Usage  openAIUsage        `json:"usage"`
+}
+
+type openAIEmbedDatum struct {
+	Object    string    `json:"object"`
+	Embedding []float32 `json:"embedding"`
+	Index     int       `json:"index"`
+}
+
+type openAIUsage struct {
+	PromptTokens int `json:"prompt_tokens"`
+	TotalTokens  int `json:"total_tokens"`
+}
+
+func (h *handler) openAIEmbeddings(w http.ResponseWriter, req bunrouter.Request) error {
+	var body openAIEmbedRequest
+	if err := json.NewDecoder(req.Body).Decode(&body); err != nil {
+		return writeJSON(w, http.StatusBadRequest, map[string]string{"error": "invalid request body"})
+	}
+
+	texts, err := toStringSlice(body.Input)
+	if err != nil {
+		return writeJSON(w, http.StatusBadRequest, map[string]string{"error": err.Error()})
+	}
+
+	client, targetName, targetURL := h.resolveClient(body.Model)
+	trace := TraceFromContext(req.Context())
+	trace.ForwardTarget = targetName
+	trace.ForwardURL = targetURL
+
+	t0 := time.Now()
+	embedResp, err := client.Embed(req.Context(), embedclient.Request{Texts: texts, Model: body.Model})
+	trace.ForwardDuration = time.Since(t0)
+	if err != nil {
+		return writeJSON(w, http.StatusBadGateway, map[string]string{"error": err.Error()})
+	}
+	trace.ForwardModel = embedResp.Model
+	trace.PromptTokens = embedResp.Usage.PromptTokens
+	trace.TotalTokens = embedResp.Usage.TotalTokens
+
+	t1 := time.Now()
+	data := make([]openAIEmbedDatum, len(embedResp.Embeddings))
+	for i, vec := range embedResp.Embeddings {
+		adapted, adaptErr := h.adapter.Adapt(vec)
+		if adaptErr != nil {
+			return writeJSON(w, http.StatusInternalServerError, map[string]string{"error": adaptErr.Error()})
+		}
+		data[i] = openAIEmbedDatum{Object: "embedding", Embedding: adapted, Index: i}
+	}
+	trace.TranslateDuration = time.Since(t1)
+
+	writeTraceHeaders(w, trace)
+
+	return writeJSON(w, http.StatusOK, openAIEmbedResponse{
+		Object: "list",
+		Data:   data,
+		Model:  embedResp.Model,
+		Usage:  openAIUsage{PromptTokens: embedResp.Usage.PromptTokens, TotalTokens: embedResp.Usage.TotalTokens},
+	})
+}
+
+// toStringSlice accepts a JSON string or array of strings.
+func toStringSlice(v interface{}) ([]string, error) {
+	switch val := v.(type) {
+	case string:
+		return []string{val}, nil
+	case []interface{}:
+		out := make([]string, len(val))
+		for i, item := range val {
+			s, ok := item.(string)
+			if !ok {
+				return nil, fmt.Errorf("input array element %d is not a string", i)
+			}
+			out[i] = s
+		}
+		return out, nil
+	default:
+		return nil, fmt.Errorf("input must be a string or array of strings")
+	}
+}