mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-22 23:03:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

s/From*Slice/From*s/ (#12255)

Browse Source
This commit is contained in:
Michael Yang
2025-10-28 12:08:49 -07:00
committed by GitHub
parent 15c7d30d9a
commit 1188f408dd
24 changed files with 95 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
model/models/mllama/model.go
View File

@@ -80,8 +80,8 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) ([]input
f32s = f32s[:m.imageSize*m.imageSize*m.numChannels*m.maxNumTiles]
}
pixelValues := ctx.Input().FromFloatSlice(f32s, m.imageSize, m.imageSize, m.numChannels, m.maxNumTiles)
aspectRatio := ctx.Input().FromIntSlice([]int32{int32(ratio.rank)}, 1)
pixelValues := ctx.Input().FromFloats(f32s, m.imageSize, m.imageSize, m.numChannels, m.maxNumTiles)
aspectRatio := ctx.Input().FromInts([]int32{int32(ratio.rank)}, 1)
positionIDs := ctx.Arange(0, 1601, 1, ml.DTypeI32)
crossAttentionStates := m.VisionModel.Forward(ctx, pixelValues, positionIDs, aspectRatio)
@@ -106,7 +106,7 @@ func (m *Model) Forward(ctx ml.Context, batch input.Batch) (ml.Tensor, error) {
crossAttentionStates = batch.Multimodal[len(batch.Multimodal)-1].Multimodal[0].Tensor
}
positions := ctx.Input().FromIntSlice(batch.Positions, len(batch.Positions))
positions := ctx.Input().FromInts(batch.Positions, len(batch.Positions))
// TODO: attention mask, cross attention mask
return m.TextModel.Forward(ctx, batch.Inputs, positions, batch.Outputs, crossAttentionStates, nil, m.Cache.(*kvcache.WrapperCache)), nil