fs: move ml.Config to fs package

model/model.go

@@ -16,7 +16,8 @@ import (
 	_ "golang.org/x/image/tiff"
 	_ "golang.org/x/image/webp"
 
-	fs "github.com/ollama/ollama/fs/ggml"
+	"github.com/ollama/ollama/fs"
+	fsggml "github.com/ollama/ollama/fs/ggml"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	_ "github.com/ollama/ollama/ml/backend"
@@ -83,10 +84,10 @@ func (m *Base) Config() config {
 	return m.config
 }
 
-var models = make(map[string]func(ml.Config) (Model, error))
+var models = make(map[string]func(fs.Config) (Model, error))
 
 // Register registers a model constructor for the given architecture
-func Register(name string, f func(ml.Config) (Model, error)) {
+func Register(name string, f func(fs.Config) (Model, error)) {
 	if _, ok := models[name]; ok {
 		panic("model: model already registered")
 	}
@@ -131,14 +132,14 @@ func NewTextProcessor(s string) (TextProcessor, error) {
 		return nil, err
 	}
 	defer r.Close()
-	meta, _, err := fs.Decode(r, -1)
+	meta, _, err := fsggml.Decode(r, -1)
 	if err != nil {
 		return nil, err
 	}
 	return getTextProcessor(meta.KV())
 }
 
-func getTextProcessor(kv fs.KV) (TextProcessor, error) {
+func getTextProcessor(kv fsggml.KV) (TextProcessor, error) {
 	arch := kv.Architecture()
 	f, ok := models[arch]
 	if !ok {

model/model_test.go

@@ -7,7 +7,8 @@ import (
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
-	fs "github.com/ollama/ollama/fs/ggml"
+	"github.com/ollama/ollama/fs"
+	fsggml "github.com/ollama/ollama/fs/ggml"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/backend/ggml"
 	"github.com/ollama/ollama/ml/nn"
@@ -139,7 +140,7 @@ func TestPopulateFieldsAlternateName(t *testing.T) {
 }
 
 func TestGetTextProcessor(t *testing.T) {
-	tp, err := getTextProcessor(fs.KV{})
+	tp, err := getTextProcessor(fsggml.KV{})
 	if err == nil {
 		t.Error("expected error")
 	} else if !strings.Contains(err.Error(), "unsupported model architecture") {
@@ -148,10 +149,10 @@ func TestGetTextProcessor(t *testing.T) {
 		t.Error("expected nil tp")
 	}
 
-	models["dummy"] = func(ml.Config) (Model, error) {
+	models["dummy"] = func(fs.Config) (Model, error) {
 		return notTextProcessorModel{}, nil
 	}
-	tp, err = getTextProcessor(fs.KV{"general.architecture": "dummy"})
+	tp, err = getTextProcessor(fsggml.KV{"general.architecture": "dummy"})
 	if err == nil {
 		t.Error("expected error")
 	} else if !strings.Contains(err.Error(), "not a TextProcessor") {

model/models/gemma2/model.go

@@ -3,6 +3,7 @@ package gemma2
 import (
 	"math"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -35,7 +36,7 @@ const (
 	gemma27BLayerCount = 46
 )
 
-func New(c ml.Config) (model.Model, error) {
+func New(c fs.Config) (model.Model, error) {
 	m := Model{
 		SentencePieceModel: model.NewSentencePieceModel(
 			&model.Vocabulary{

model/models/gemma3/model.go

@@ -6,6 +6,7 @@ import (
 	"math"
 	"slices"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -52,7 +53,7 @@ func (p *MultiModalProjector) Forward(ctx ml.Context, visionOutputs ml.Tensor, i
 	return visionOutputs
 }
 
-func New(c ml.Config) (model.Model, error) {
+func New(c fs.Config) (model.Model, error) {
 	m := Model{
 		SentencePieceModel: model.NewSentencePieceModel(
 			&model.Vocabulary{

model/models/gemma3/model_text.go

@@ -3,6 +3,7 @@ package gemma3
 import (
 	"math"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -40,7 +41,7 @@ const (
 	cacheTypeCausal
 )
 
-func newTextModel(c ml.Config) *TextModel {
+func newTextModel(c fs.Config) *TextModel {
 	numBlocks := int(c.Uint("block_count"))
 
 	m := TextModel{

model/models/gemma3/model_vision.go

@@ -3,6 +3,7 @@ package gemma3
 import (
 	"math"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
 )
@@ -111,7 +112,7 @@ func (m *VisionModel) Forward(ctx ml.Context, pixelValues ml.Tensor) ml.Tensor {
 	return hiddenState
 }
 
-func newVisionModel(c ml.Config) *VisionModel {
+func newVisionModel(c fs.Config) *VisionModel {
 	return &VisionModel{
 		Layers: make([]VisionEncoderLayer, c.Uint("vision.block_count")),
 		VisionModelOptions: &VisionModelOptions{

model/models/gemma3/process_image.go

@@ -3,7 +3,7 @@ package gemma3
 import (
 	"image"
 
-	"github.com/ollama/ollama/ml"
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/model/imageproc"
 )
 
@@ -11,7 +11,7 @@ type ImageProcessor struct {
 	imageSize, patchSize, numChannels int
 }
 
-func newImageProcessor(c ml.Config) ImageProcessor {
+func newImageProcessor(c fs.Config) ImageProcessor {
 	return ImageProcessor{
 		imageSize:   int(c.Uint("vision.image_size")),
 		patchSize:   int(c.Uint("vision.patch_size")),

model/models/llama/model.go

@@ -5,6 +5,7 @@ import (
 	"math"
 	"strings"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -30,7 +31,7 @@ type Model struct {
 	*Options
 }
 
-func New(c ml.Config) (model.Model, error) {
+func New(c fs.Config) (model.Model, error) {
 	if !strings.EqualFold(c.String("tokenizer.ggml.model"), "gpt2") {
 		return nil, fmt.Errorf("tokenizer %s not yet supported", c.String("tokenizer.ggml.model"))
 	}

model/models/mllama/model.go

@@ -8,6 +8,7 @@ import (
 	"image"
 	"slices"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -32,7 +33,7 @@ const (
 	selfAttentionLayer
 )
 
-func New(c ml.Config) (model.Model, error) {
+func New(c fs.Config) (model.Model, error) {
 	// Verify unified config
 	if c.Uint("vision.block_count") == 0 {
 		return nil, fmt.Errorf("non-unified vision model not supported")

model/models/mllama/model_text.go

@@ -4,6 +4,7 @@ import (
 	"math"
 	"slices"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
@@ -220,7 +221,7 @@ func (m *TextModel) Forward(ctx ml.Context, inputIDs, positionIDs, outputs, mask
 	return m.Output.Forward(ctx, hiddenState)
 }
 
-func newTextModel(c ml.Config) *TextModel {
+func newTextModel(c fs.Config) *TextModel {
 	var decoderLayers []TextDecoderLayer
 	for i := range c.Uint("block_count") {
 		var textDecoderLayer TextDecoderLayer

model/models/mllama/model_vision.go

@@ -4,6 +4,7 @@ import (
 	"math"
 	"slices"
 
+	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
 )
@@ -213,7 +214,7 @@ func (m *VisionModel) Forward(ctx ml.Context, pixelValues, positionIDs, aspectRa
 	return hiddenState.Concat(ctx, hiddenStates, 0)
 }
 
-func newVisionModel(c ml.Config) *VisionModel {
+func newVisionModel(c fs.Config) *VisionModel {
 	return &VisionModel{
 		Transformer:       &VisionEncoder{Layers: make([]VisionEncoderLayer, c.Uint("vision.block_count"))},
 		GlobalTransformer: &VisionEncoder{Layers: make([]VisionEncoderLayer, c.Uint("vision.global.block_count"))},

model/models/mllama/process_image.go

@@ -8,14 +8,14 @@ import (
 
 	"golang.org/x/image/draw"
 
-	"github.com/ollama/ollama/ml"
+	"github.com/ollama/ollama/fs"
 )
 
 type ImageProcessor struct {
 	imageSize, numChannels, maxNumTiles int
 }
 
-func newImageProcessor(c ml.Config) ImageProcessor {
+func newImageProcessor(c fs.Config) ImageProcessor {
 	return ImageProcessor{
 		imageSize:   int(c.Uint("vision.image_size")),
 		numChannels: int(c.Uint("vision.num_channels")),