refactor rope

change to a flatter directory structure and group the options with the
function

update models to call rope in one place

model/models/mistral3/model_text.go

@@ -8,7 +8,6 @@ import (
 	"github.com/ollama/ollama/kvcache"
 	"github.com/ollama/ollama/ml"
 	"github.com/ollama/ollama/ml/nn"
-	"github.com/ollama/ollama/ml/nn/fast"
 	"github.com/ollama/ollama/model/input"
 )
 
@@ -20,6 +19,10 @@ type TextOptions struct {
 	ropeScalingBeta                  float32
 }
 
+func (o TextOptions) applyRotaryPositionEmbeddings(ctx ml.Context, states, positions ml.Tensor) ml.Tensor {
+	return nn.RoPE(ctx, states, positions, o.ropeDim, o.ropeBase, 1./o.ropeScale)
+}
+
 type TextModel struct {
 	TokenEmbedding *nn.Embedding `gguf:"token_embd"`
 	Layers         []Layer       `gguf:"blk"`
@@ -42,11 +45,11 @@ func (sa *SelfAttention) Forward(ctx ml.Context, hiddenState, positionIDs, posit
 
 	q := sa.Query.Forward(ctx, hiddenState)
 	q = q.Reshape(ctx, headDim, opts.numHeads, batchSize)
-	q = fast.RoPE(ctx, q, positionIDs, opts.ropeDim, opts.ropeBase, 1./opts.ropeScale)
+	q = opts.applyRotaryPositionEmbeddings(ctx, q, positionIDs)
 
 	k := sa.Key.Forward(ctx, hiddenState)
 	k = k.Reshape(ctx, headDim, opts.numKVHeads, batchSize)
-	k = fast.RoPE(ctx, k, positionIDs, opts.ropeDim, opts.ropeBase, 1./opts.ropeScale)
+	k = opts.applyRotaryPositionEmbeddings(ctx, k, positionIDs)
 
 	v := sa.Value.Forward(ctx, hiddenState)
 	v = v.Reshape(ctx, headDim, opts.numKVHeads, batchSize)
@@ -61,7 +64,7 @@ func (sa *SelfAttention) Forward(ctx ml.Context, hiddenState, positionIDs, posit
 }
 
 func (m *TextModel) Shift(ctx ml.Context, layer int, key, shift ml.Tensor) (ml.Tensor, error) {
-	return fast.RoPE(ctx, key, shift, m.ropeDim, m.ropeBase, 1./m.ropeScale), nil
+	return m.applyRotaryPositionEmbeddings(ctx, key, shift), nil
 }
 
 type MLP struct {