matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 08:17:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

update model

Browse Source
This commit is contained in:
Michael Yang
2025-03-08 12:31:57 -08:00
parent 8934324b72
commit 46bb0169c4
5 changed files with 28 additions and 74 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
model/models/gemma3/model_text.go
View File

@@ -169,14 +169,14 @@ func (l *TextLayer) Forward(ctx ml.Context, layer int, hiddenState, positionIDs,
func (m *TextModel) Forward(ctx ml.Context, inputs, positions, outputs ml.Tensor, multimodal []input.MultimodalIndex, cache kvcache.Cache) ml.Tensor {
hiddenState := m.TokenEmbedding.Forward(ctx, inputs)
hiddenState = hiddenState.Scale(ctx, math.Sqrt(float64(m.TextOptions.hiddenSize)))
if multimodal != nil {
visionOutputs := multimodal[0].Multimodal.(ml.Tensor)
offset := multimodal[0].Index - 1 - visionOutputs.Dim(1)
hiddenState = hiddenState.Set(ctx, visionOutputs, offset*hiddenState.Stride(0))
}
hiddenState = hiddenState.Scale(ctx, math.Sqrt(float64(m.TextOptions.hiddenSize)))
if len(m.Layers) == gemma27BLayerCount {
m.TextOptions.largeModelScaling = true
}