mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-23 23:18:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

chore: update models to use slice/chunk/chunksections (#12934)

Browse Source 
* use slice/chunks

* bert

* llama4

* gemma3n

* gptoss

* mistral3

* qwen3vl

* qwen25vl

* deepseek2

* remove unused ops
This commit is contained in:
Michael Yang
2025-11-13 15:20:12 -08:00
committed by GitHub
parent c114987523
commit 333203d871
13 changed files with 59 additions and 135 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
model/models/llama4/model.go
View File

@@ -105,9 +105,7 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) ([]input
for range aspectRatio.Y {
for x := range aspectRatio.X {
view := projectedOutputs.View(ctx, projectedOutputs.Stride(1)*offset,
projectedOutputs.Dim(0), projectedOutputs.Stride(1),
patchesPerChunk)
view := projectedOutputs.Slice(ctx, 1, offset, offset+patchesPerChunk, 1)
var separator separator
if x < aspectRatio.X-1 {
separator.x = true // <|tile_x_separator|>
@@ -120,9 +118,7 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) ([]input
}
}
view := projectedOutputs.View(ctx, projectedOutputs.Stride(1)*offset,
projectedOutputs.Dim(0), projectedOutputs.Stride(1),
patchesPerChunk)
view := projectedOutputs.Slice(ctx, 1, offset, offset+patchesPerChunk, 1)
multimodal = append(multimodal, input.Multimodal{Tensor: view, Data: &separator{}})
return multimodal, nil