gemma3: scale in attention

model/models/gemma2/model.go

@@ -69,10 +69,10 @@ func New(c fs.Config) (model.Model, error) {
 		},
 	}
 
-	slidingWindowLen := int32(c.Uint("attention.sliding_window"))
+	m.Cache = kvcache.NewWrapperCache(
-	m.Cache = kvcache.NewWrapperCache(kvcache.NewSWACache(slidingWindowLen, m.Shift), kvcache.NewCausalCache(m.Shift))
+		kvcache.NewSWACache(int32(c.Uint("attention.sliding_window")), m.Shift),
-	m.Cache.SetConfig(ml.CacheConfig{})
+		kvcache.NewCausalCache(m.Shift),
-
+	)
 	return &m, nil
 }
 
@@ -90,12 +90,6 @@ func (sa *SelfAttention) Forward(ctx ml.Context, hiddenState, positionIDs ml.Ten
 	q = q.Reshape(ctx, opts.attnKeyLen, opts.numHeads, batchSize)
 	q = fast.RoPE(ctx, q, positionIDs, opts.attnKeyLen, opts.ropeBase, opts.ropeScale, rope.WithTypeNeoX())
 
-	if opts.largeModelScaling {
-		q = q.Scale(ctx, 1.0/math.Sqrt(float64(opts.hiddenSize/opts.numHeads)))
-	} else {
-		q = q.Scale(ctx, 1.0/math.Sqrt(float64(opts.attnKeyLen)))
-	}
-
 	k := sa.Key.Forward(ctx, hiddenState)
 	k = k.Reshape(ctx, opts.attnKeyLen, opts.numKVHeads, batchSize)
 	k = fast.RoPE(ctx, k, positionIDs, opts.attnKeyLen, opts.ropeBase, opts.ropeScale, rope.WithTypeNeoX())
@@ -103,28 +97,14 @@ func (sa *SelfAttention) Forward(ctx ml.Context, hiddenState, positionIDs ml.Ten
 	v := sa.Value.Forward(ctx, hiddenState)
 	v = v.Reshape(ctx, opts.attnValLen, opts.numKVHeads, batchSize)
 
-	cache.Put(ctx, k, v)
+	scale := 1.0 / math.Sqrt(float64(opts.attnKeyLen))
-	k, v, mask := cache.Get(ctx)
+	if opts.largeModelScaling {
+		scale = 1.0 / math.Sqrt(float64(opts.hiddenSize/opts.numHeads))
+	}
 
-	q = q.Permute(ctx, 0, 2, 1, 3)
+	attn := nn.Attention(ctx, q, k, v, scale, cache)
-	k = k.Permute(ctx, 0, 2, 1, 3)
+	attn = attn.Reshape(ctx, opts.attnValLen*opts.numHeads, batchSize)
-	v = v.Permute(ctx, 1, 2, 0, 3).Contiguous(ctx)
+	return sa.Output.Forward(ctx, attn)
-
-	kq := k.Mulmat(ctx, q)
-
-	// logit softcap
-	kq = kq.Scale(ctx, 1.0/float64(opts.attnLogitSoftcap))
-	kq = kq.Tanh(ctx)
-	kq = kq.Scale(ctx, float64(opts.attnLogitSoftcap))
-
-	kq = kq.Add(ctx, mask)
-	kq = kq.Softmax(ctx)
-
-	kqv := v.Mulmat(ctx, kq)
-	kqv = kqv.Permute(ctx, 0, 2, 1, 3).Contiguous(ctx)
-	kqv = kqv.Reshape(ctx, opts.attnValLen*opts.numHeads, batchSize)
-
-	return sa.Output.Forward(ctx, kqv)
 }
 
 func (m *Model) Shift(ctx ml.Context, layer int, key, shift ml.Tensor) (ml.Tensor, error) {

model/models/gemma3/model_text.go

@@ -86,12 +86,6 @@ func (sa *TextSelfAttention) Forward(ctx ml.Context, layer int, hiddenState, pos
 	q = sa.QueryNorm.Forward(ctx, q, opts.eps)
 	q = fast.RoPE(ctx, q, positionIDs, opts.attnKeyLen, ropeBase, opts.ropeScale, rope.WithTypeNeoX())
 
-	if opts.largeModelScaling {
-		q = q.Scale(ctx, 1.0/math.Sqrt(float64(opts.hiddenSize/opts.numHeads)))
-	} else {
-		q = q.Scale(ctx, 1.0/math.Sqrt(float64(opts.attnKeyLen)))
-	}
-
 	k := sa.Key.Forward(ctx, hiddenState)
 	k = k.Reshape(ctx, opts.attnKeyLen, opts.numKVHeads, batchSize)
 	k = sa.KeyNorm.Forward(ctx, k, opts.eps)
@@ -100,8 +94,12 @@ func (sa *TextSelfAttention) Forward(ctx ml.Context, layer int, hiddenState, pos
 	v := sa.Value.Forward(ctx, hiddenState)
 	v = v.Reshape(ctx, opts.attnValLen, opts.numKVHeads, batchSize)
 
-	scaleFactor := 1.0
+	scale := 1.0 / math.Sqrt(float64(opts.attnKeyLen))
-	kqv := nn.Attention(ctx, q, k, v, scaleFactor, cache)
+	if opts.largeModelScaling {
+		scale = 1.0 / math.Sqrt(float64(opts.hiddenSize/opts.numHeads))
+	}
+
+	kqv := nn.Attention(ctx, q, k, v, scale, cache)
 	kqv = kqv.Reshape(ctx, opts.attnValLen*opts.numHeads, batchSize)
 
 	return sa.Output.Forward(ctx, kqv)