highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-11-12 08:57:18 +01:00
Code Issues Packages Projects Releases Wiki Activity

add pre:, suf: to tags (#12274)

Browse Source
This commit is contained in:
Michael Yang
2025-09-23 16:08:57 -07:00
committed by GitHub
parent a40d427bce
commit bf78ed6ee9
3 changed files with 101 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
model/models/llama4/model_text.go
View File

@@ -88,22 +88,10 @@ func (e *TextExperts) Forward(ctx ml.Context, hiddenStates, routerLogits ml.Tens
return nextStates
}
// TextSharedExpert is TextMLP with different tensor names
type TextSharedExpert struct {
Gate *nn.Linear `gguf:"ffn_gate_shexp"`
Up *nn.Linear `gguf:"ffn_up_shexp"`
Down *nn.Linear `gguf:"ffn_down_shexp"`
}
func (mlp *TextSharedExpert) Forward(ctx ml.Context, hiddenStates ml.Tensor, opts *TextOptions) ml.Tensor {
hiddenStates = mlp.Gate.Forward(ctx, hiddenStates).SILU(ctx, mlp.Up.Forward(ctx, hiddenStates))
return mlp.Down.Forward(ctx, hiddenStates)
}
type TextMOE struct {
Router *nn.Linear `gguf:"ffn_gate_inp"`
Experts *TextExperts
SharedExpert *TextSharedExpert
SharedExpert *TextMLP `gguf:",suf:_shexp"`
}
func (moe *TextMOE) Forward(ctx ml.Context, hiddenStates ml.Tensor, opts *TextOptions) ml.Tensor {