mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 14:26:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

contiguous input per layer (#12686)

Browse Source 
Co-authored-by: Michael Yang <git@mxy.ng>
This commit is contained in:
Daniel Hiltgen
2025-10-17 18:39:18 -07:00
committed by GitHub
parent ba2253dc30
commit bc1a818fdc
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
model/models/gemma3n/model_text.go
View File

@@ -65,7 +65,7 @@ func (m *TextModel) Forward(ctx ml.Context, batch input.Batch, cache kvcache.Cac
cache.(*kvcache.WrapperCache).SetLayerType(layerType)
// inputPerLayer = inputsPerLayer[:, i, :]
inputPerLayer := inputsPerLayer.View(ctx, i*inputsPerLayer.Stride(1), inputsPerLayer.Dim(0), inputsPerLayer.Stride(2), inputsPerLayer.Dim(2))
inputPerLayer := inputsPerLayer.View(ctx, i*inputsPerLayer.Stride(1), inputsPerLayer.Dim(0), inputsPerLayer.Stride(2), inputsPerLayer.Dim(2)).Contiguous(ctx)
hiddenStates = layer.Forward(ctx, hiddenStates, inputPerLayer, positions, one, cache, i >= firstSharedKeyValue, ropeBase, float64(m.activationSparsityScale[i]), &m.TextOptions)
}