mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

batch: use tensors for outputs (#12185)

Browse Source 
this cleans up the model interface slightly without too much impact in
other areas
This commit is contained in:
Michael Yang
2025-09-15 14:33:06 -07:00
committed by GitHub
parent 92b96d54ef
commit 6f7117145f
14 changed files with 27 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
model/models/gemma3/model_text.go
View File

@@ -161,7 +161,6 @@ func (l *TextLayer) Forward(ctx ml.Context, layer int, hiddenState, positionIDs,
func (m *TextModel) Forward(ctx ml.Context, batch input.Batch, cache kvcache.Cache) ml.Tensor {
positions := ctx.Input().FromIntSlice(batch.Positions, len(batch.Positions))
outputs := ctx.Input().FromIntSlice(batch.Outputs, len(batch.Outputs))
hiddenState := m.TokenEmbedding.Forward(ctx, batch.Inputs)
hiddenState = hiddenState.Scale(ctx, math.Sqrt(float64(m.TextConfig.hiddenSize)))
@@ -194,7 +193,7 @@ func (m *TextModel) Forward(ctx ml.Context, batch input.Batch, cache kvcache.Cac
var lastLayerOutputs ml.Tensor
if i == len(m.Layers)-1 {
lastLayerOutputs = outputs
lastLayerOutputs = batch.Outputs
}
hiddenState = layer.Forward(ctx, i, hiddenState, positions, lastLayerOutputs, cache, m.TextConfig)