mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-25 07:58:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

use split activations when possible (#12293)

Browse Source 
* use ggml_*_split activations when possible

* forward qkv
This commit is contained in:
Michael Yang
2025-09-16 09:51:19 -07:00
committed by GitHub
parent c253433d68
commit ad95d5b30b
16 changed files with 59 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
model/models/mistral3/model_text.go
View File

@@ -65,7 +65,7 @@ type MLP struct {
}
func (mlp *MLP) Forward(ctx ml.Context, hiddenState ml.Tensor, opts *TextOptions) ml.Tensor {
hiddenState = mlp.Gate.Forward(ctx, hiddenState).SILU(ctx).Mul(ctx, mlp.Up.Forward(ctx, hiddenState))
hiddenState = mlp.Gate.Forward(ctx, hiddenState).SILU(ctx, mlp.Up.Forward(ctx, hiddenState))
return mlp.Down.Forward(ctx, hiddenState)
}

2
model/models/mistral3/model_vision.go
View File

@@ -51,7 +51,7 @@ type VisionMLP struct {
}
func (mlp *VisionMLP) Forward(ctx ml.Context, hiddenStates ml.Tensor, opts *VisionModelOptions) ml.Tensor {
hiddenStates = mlp.Gate.Forward(ctx, hiddenStates).SILU(ctx).Mul(ctx, mlp.Up.Forward(ctx, hiddenStates))
hiddenStates = mlp.Gate.Forward(ctx, hiddenStates).SILU(ctx, mlp.Up.Forward(ctx, hiddenStates))
return mlp.Down.Forward(ctx, hiddenStates)
}