model.go

package olmo3

import (
	"fmt"
	"math"

	"github.com/ollama/ollama/fs"
	"github.com/ollama/ollama/kvcache"
	"github.com/ollama/ollama/ml"
	"github.com/ollama/ollama/ml/nn"
	"github.com/ollama/ollama/ml/nn/rope"
	"github.com/ollama/ollama/model"
	"github.com/ollama/ollama/model/input"
)

const (
	cacheTypeSWA    = 0
	cacheTypeCausal = 1
)

type Options struct {
	hiddenSize, numHeads, numKVHeads int
	eps, ropeBase, ropeScale         float32

	originalContextLength int
	attnFactor            float32

	ropeType          string
	ropeExtrapolation float32

	slidingWindowPattern []bool
}

type Model struct {
	model.Base
	model.TextProcessor

	TokenEmbedding *nn.Embedding `gguf:"token_embd"`
	Layers         []Layer       `gguf:"blk"`
	OutputNorm     *nn.RMSNorm   `gguf:"output_norm"`
	Output         *nn.Linear    `gguf:"output,alt:token_embd"`

	Options
}

func New(c fs.Config) (model.Model, error) {
	vocabulary := model.Vocabulary{
		Values: c.Strings("tokenizer.ggml.tokens"),
		Scores: c.Floats("tokenizer.ggml.scores"),
		Types:  c.Ints("tokenizer.ggml.token_type"),
		Merges: c.Strings("tokenizer.ggml.merges"),
		AddBOS: c.Bool("tokenizer.ggml.add_bos_token", false),
		BOS:    []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
		AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
		EOS: append(
			[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
			c.Ints("tokenizer.ggml.eos_token_ids")...,
		),
	}

	processor := model.NewBytePairEncoding(
		&vocabulary,
		"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
	)

	m := Model{
		TextProcessor: processor,
		Layers:        make([]Layer, c.Uint("block_count")),
		Options: Options{
			hiddenSize:            int(c.Uint("embedding_length")),
			numHeads:              int(c.Uint("attention.head_count")),
			numKVHeads:            int(c.Uint("attention.head_count_kv")),
			eps:                   c.Float("attention.layer_norm_rms_epsilon"),
			ropeBase:              c.Float("rope.freq_base", 1e4),
			ropeScale:             c.Float("rope.scaling.factor", 1),
			originalContextLength: int(c.Uint("rope.scaling.original_context_length")),
			attnFactor:            c.Float("rope.scaling.attn_factor", 1),
			ropeType:              c.String("rope.scaling.type"),
			ropeExtrapolation:     c.Float("rope.scaling.extrapolation_factor", 1.0),
			slidingWindowPattern:  c.Bools("attention.sliding_window_pattern"),
		},
	}

	m.Cache = kvcache.NewWrapperCache(
		kvcache.NewSWACache(int32(c.Uint("attention.sliding_window")), m.Shift),
		kvcache.NewCausalCache(m.Shift),
	)

	return &m, nil
}

type SelfAttention struct {
	Query  *nn.Linear  `gguf:"attn_q"`
	Key    *nn.Linear  `gguf:"attn_k"`
	Value  *nn.Linear  `gguf:"attn_v"`
	Output *nn.Linear  `gguf:"attn_output"`
	QNorm  *nn.RMSNorm `gguf:"attn_q_norm"`
	KNorm  *nn.RMSNorm `gguf:"attn_k_norm"`
}

func (o Options) applyRotaryPositionEmbeddings(ctx ml.Context, states, positions ml.Tensor, isSWA bool) ml.Tensor {
	freqScale := float32(1.0)
	ropeOpts := []func(*rope.Options){rope.WithTypeNeoX()}

	if !isSWA {
		freqScale = 1. / o.ropeScale
		if o.originalContextLength > 0 {
			ropeOpts = append(ropeOpts,
				rope.WithOriginalContextLength(o.originalContextLength),
				rope.WithExtrapolationFactor(o.ropeExtrapolation),
			)
		}
	}

	return nn.RoPE(ctx, states, positions, o.hiddenSize/o.numHeads, o.ropeBase, freqScale, ropeOpts...)
}

func (sa *SelfAttention) Forward(ctx ml.Context, hiddenState, positions ml.Tensor, cache kvcache.Cache, m *Model, isSWA bool) ml.Tensor {
	batchSize := hiddenState.Dim(1)
	headDim := m.hiddenSize / m.numHeads

	query := sa.Query.Forward(ctx, hiddenState)
	query = sa.QNorm.Forward(ctx, query, m.eps)
	query = query.Reshape(ctx, headDim, m.numHeads, batchSize)
	query = m.Options.applyRotaryPositionEmbeddings(ctx, query, positions, isSWA)

	key := sa.Key.Forward(ctx, hiddenState)
	key = sa.KNorm.Forward(ctx, key, m.eps)
	key = key.Reshape(ctx, headDim, m.numKVHeads, batchSize)
	key = m.Options.applyRotaryPositionEmbeddings(ctx, key, positions, isSWA)

	value := sa.Value.Forward(ctx, hiddenState)
	value = value.Reshape(ctx, headDim, m.numKVHeads, batchSize)

	attention := nn.Attention(ctx, query, key, value, 1.0/math.Sqrt(float64(headDim)), cache)
	attention = attention.Reshape(ctx, m.hiddenSize, batchSize)

	return sa.Output.Forward(ctx, attention)
}

func (m *Model) Shift(ctx ml.Context, layer int, key, shift ml.Tensor) (ml.Tensor, error) {
	isSWA := m.isSWALayer(layer)
	return m.Options.applyRotaryPositionEmbeddings(ctx, key, shift, isSWA), nil
}

type MLP struct {
	Up   *nn.Linear `gguf:"ffn_up"`
	Down *nn.Linear `gguf:"ffn_down"`
	Gate *nn.Linear `gguf:"ffn_gate"`
}

func (mlp *MLP) Forward(ctx ml.Context, hiddenState ml.Tensor, m *Model) ml.Tensor {
	hiddenState = mlp.Gate.Forward(ctx, hiddenState).SILU(ctx, mlp.Up.Forward(ctx, hiddenState))
	return mlp.Down.Forward(ctx, hiddenState)
}

type Layer struct {
	SelfAttention     *SelfAttention
	PostAttentionNorm *nn.RMSNorm `gguf:"post_attention_norm"`
	MLP               *MLP
	PostFFWNorm       *nn.RMSNorm `gguf:"post_ffw_norm"`
}

func (l *Layer) Forward(ctx ml.Context, hiddenState, positions, outputs ml.Tensor, cache kvcache.Cache, m *Model, isSWA bool) ml.Tensor {
	residual := hiddenState

	hiddenState = l.SelfAttention.Forward(ctx, hiddenState, positions, cache, m, isSWA)

	if outputs != nil {
		hiddenState = hiddenState.Rows(ctx, outputs)
		residual = residual.Rows(ctx, outputs)
	}
	hiddenState = l.PostAttentionNorm.Forward(ctx, hiddenState, m.eps)

	hiddenState = hiddenState.Add(ctx, residual)
	residual = hiddenState

	hiddenState = l.MLP.Forward(ctx, hiddenState, m)
	hiddenState = l.PostFFWNorm.Forward(ctx, hiddenState, m.eps)

	return hiddenState.Add(ctx, residual)
}

// OLMo3 has Sliding Window Attention (SWA) for 3 out of every 4 layers.
func (m *Model) isSWALayer(layerIdx int) bool {
	return m.Options.slidingWindowPattern[layerIdx]
}

func (m *Model) Forward(ctx ml.Context, batch input.Batch) (ml.Tensor, error) {
	positions := ctx.Input().FromInts(batch.Positions, len(batch.Positions))

	hiddenState := m.TokenEmbedding.Forward(ctx, batch.Inputs)

	for i, layer := range m.Layers {
		m.Cache.SetLayer(i)
		cacheType := cacheTypeSWA

		isSWA := m.isSWALayer(i)
		if !isSWA {
			cacheType = cacheTypeCausal
		}

		wc, ok := m.Cache.(*kvcache.WrapperCache)
		if !ok {
			return nil, fmt.Errorf("expected *kvcache.WrapperCache, got %T", m.Cache)
		}
		wc.SetLayerType(cacheType)

		var outputs ml.Tensor
		if i == len(m.Layers)-1 {
			outputs = batch.Outputs
		}

		hiddenState = layer.Forward(ctx, hiddenState, positions, outputs, m.Cache, m, isSWA)
	}

	hiddenState = m.OutputNorm.Forward(ctx, hiddenState, m.eps)
	return m.Output.Forward(ctx, hiddenState), nil
}

func init() {
	model.Register("olmo3", New)
}