matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 07:46:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

set non-causal attention

Browse Source
This commit is contained in:
Michael Yang
2025-03-07 13:52:45 -08:00
parent 631fecc6d9
commit 0df1800436
6 changed files with 57 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
ml/backend.go
View File

@@ -148,6 +148,7 @@ type Tensor interface {
View(ctx Context, offset int, shape ...int) Tensor
Permute(ctx Context, shape ...int) Tensor
Contiguous(ctx Context) Tensor
Set(ctx Context, t2 Tensor, offset int, strides ...int) Tensor
Pad(ctx Context, shape ...int) Tensor
Unpad(ctx Context, shape ...int) Tensor

14
ml/backend/ggml/ggml.go
View File

@@ -954,6 +954,20 @@ func (t *Tensor) AvgPool1D(ctx ml.Context, k, s, p int) ml.Tensor {
}
}
func (t *Tensor) Set(ctx ml.Context, t2 ml.Tensor, offset int, strides ...int) ml.Tensor {
var tt *C.struct_ggml_tensor
switch len(strides) {
case 0:
tt = C.ggml_set_1d_inplace(ctx.(*Context).ctx, t.t, t2.(*Tensor).t, C.size_t(offset))
case 1:
tt = C.ggml_set_2d_inplace(ctx.(*Context).ctx, t.t, t2.(*Tensor).t, C.size_t(offset), C.size_t(strides[0]))
default:
panic("unsupported number of dimensions")
}
return &Tensor{b: t.b, t: tt}
}
func (t *Tensor) ScaledDotProductAttention(ctx ml.Context, key, value, mask ml.Tensor, scale float64) ml.Tensor {
var kqMask *C.struct_ggml_tensor
if mask != nil {