Merge pull request #4502 from ollama/mxyng/fix-quantize

fix quantize file types

server/images.go

@@ -341,8 +341,8 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 				}
 			} else if strings.HasPrefix(c.Args, "@") {
 				digest := strings.TrimPrefix(c.Args, "@")
-				if ib, ok := intermediateBlobs.Load(digest); ok {
-					p, err := GetBlobsPath(ib.(string))
+				if ib, ok := intermediateBlobs[digest]; ok {
+					p, err := GetBlobsPath(ib)
 					if err != nil {
 						return err
 					}
@@ -352,8 +352,8 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 					} else if err != nil {
 						return err
 					} else {
-						fn(api.ProgressResponse{Status: fmt.Sprintf("using cached layer %s", ib.(string))})
-						digest = ib.(string)
+						fn(api.ProgressResponse{Status: fmt.Sprintf("using cached layer %s", ib)})
+						digest = ib
 					}
 				}
 
@@ -415,14 +415,17 @@ func CreateModel(ctx context.Context, name, modelFileDir, quantization string, m
 							return err
 						}
 
-						f16digest := baseLayer.Layer.Digest
-
-						baseLayer.Layer, err = NewLayer(temp, baseLayer.Layer.MediaType)
+						layers, err := parseFromFile(ctx, temp, "", fn)
 						if err != nil {
 							return err
 						}
 
-						intermediateBlobs.Store(f16digest, baseLayer.Layer.Digest)
+						if len(layers) != 1 {
+							return errors.New("quantization failed")
+						}
+
+						baseLayer.Layer = layers[0].Layer
+						baseLayer.GGML = layers[0].GGML
 					}
 				}
 

server/model.go

@@ -10,7 +10,6 @@ import (
 	"net/http"
 	"os"
 	"path/filepath"
-	"sync"
 
 	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/convert"
@@ -18,7 +17,7 @@ import (
 	"github.com/ollama/ollama/types/model"
 )
 
-var intermediateBlobs sync.Map
+var intermediateBlobs map[string]string = make(map[string]string)
 
 type layerWithGGML struct {
 	*Layer
@@ -169,7 +168,7 @@ func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(a
 
 	layer, err := NewLayer(temp, "application/vnd.ollama.image.model")
 	if err != nil {
-		return nil, fmt.Errorf("aaa: %w", err)
+		return nil, err
 	}
 
 	bin, err := layer.Open()
@@ -185,7 +184,7 @@ func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(a
 
 	layers = append(layers, &layerWithGGML{layer, ggml})
 
-	intermediateBlobs.Store(digest, layer.Digest)
+	intermediateBlobs[digest] = layer.Digest
 	return layers, nil
 }
 

server/routes.go

@@ -841,16 +841,16 @@ func (s *Server) HeadBlobHandler(c *gin.Context) {
 }
 
 func (s *Server) CreateBlobHandler(c *gin.Context) {
-	ib, ok := intermediateBlobs.Load(c.Param("digest"))
-	if ok {
-		p, err := GetBlobsPath(ib.(string))
+	if ib, ok := intermediateBlobs[c.Param("digest")]; ok {
+		p, err := GetBlobsPath(ib)
 		if err != nil {
 			c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 			return
 		}
 
 		if _, err := os.Stat(p); errors.Is(err, os.ErrNotExist) {
-			intermediateBlobs.Delete(c.Param("digest"))
+			slog.Info("evicting intermediate blob which no longer exists", "digest", ib)
+			delete(intermediateBlobs, c.Param("digest"))
 		} else if err != nil {
 			c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 			return