feat: add dimensions field to embed requests (#12242)

* feat: add field to truncate embeddings

* add openai embeddings for dimensions

api/types.go

@@ -388,8 +388,12 @@ type EmbedRequest struct {
 	// this request.
 	KeepAlive *Duration `json:"keep_alive,omitempty"`
 
+	// Truncate truncates the input to fit the model's max sequence length.
 	Truncate *bool `json:"truncate,omitempty"`
 
+	// Dimensions truncates the output embedding to the specified dimension.
+	Dimensions int `json:"dimensions,omitempty"`
+
 	// Options lists model-specific options.
 	Options map[string]any `json:"options"`
 }

docs/api.md

@@ -1708,6 +1708,7 @@ Advanced parameters:
 - `truncate`: truncates the end of each input to fit within context length. Returns error if `false` and context length is exceeded. Defaults to `true`
 - `options`: additional model parameters listed in the documentation for the [Modelfile](./modelfile.md#valid-parameters-and-values) such as `temperature`
 - `keep_alive`: controls how long the model will stay loaded into memory following the request (default: `5m`)
+- `dimensions`: number of dimensions for the embedding
 
 ### Examples
 

openai/openai.go

@@ -78,6 +78,7 @@ type JsonSchema struct {
 type EmbedRequest struct {
 	Input      any    `json:"input"`
 	Model      string `json:"model"`
+	Dimensions int    `json:"dimensions,omitempty"`
 }
 
 type StreamOptions struct {
@@ -1005,7 +1006,7 @@ func EmbeddingsMiddleware() gin.HandlerFunc {
 		}
 
 		var b bytes.Buffer
-		if err := json.NewEncoder(&b).Encode(api.EmbedRequest{Model: req.Model, Input: req.Input}); err != nil {
+		if err := json.NewEncoder(&b).Encode(api.EmbedRequest{Model: req.Model, Input: req.Input, Dimensions: req.Dimensions}); err != nil {
 			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
 			return
 		}

server/routes.go

@@ -558,7 +558,12 @@ func (s *Server) EmbedHandler(c *gin.Context) {
 			if err != nil {
 				return err
 			}
-			embeddings[i] = normalize(embedding)
+			// TODO: this first normalization should be done by the model
+			embedding = normalize(embedding)
+			if req.Dimensions > 0 && req.Dimensions < len(embedding) {
+				embedding = normalize(embedding[:req.Dimensions])
+			}
+			embeddings[i] = embedding
 			return nil
 		})
 	}
@@ -584,11 +589,7 @@ func normalize(vec []float32) []float32 {
 		sum += v * v
 	}
 
-	norm := float32(0.0)
-	if sum > 0 {
-		norm = float32(1.0 / math.Sqrt(float64(sum)))
-	}
-
+	norm := float32(1.0 / max(math.Sqrt(float64(sum)), 1e-12))
 	for i := range vec {
 		vec[i] *= norm
 	}