Skip to main content

GgufCacheGeometry

Fixed recurrent state and per-token attention dimensions of one artifact.

attentionLayersAttentionlayers (integer)required

Target full-attention layers.

Possible values: >= 0

recurrentLayersRecurrentlayers (integer)required

Target recurrent layers.

Possible values: >= 0

nextnLayersNextnlayers (integer)required

Embedded MTP attention layers.

Possible values: >= 0

keyWidthKeywidth (integer)required

Key elements per token per layer.

Possible values: > 0

valueWidthValuewidth (integer)required

Value elements per token per layer.

Possible values: > 0

convolutionWidthConvolutionwidth (integer)required

Convolution-state elements per recurrent layer and row.

Possible values: >= 0

recurrentWidthRecurrentwidth (integer)required

Recurrent-state elements per recurrent layer and row.

Possible values: >= 0

GgufCacheGeometry
{
"attentionLayers": 0,
"recurrentLayers": 0,
"nextnLayers": 0,
"keyWidth": 0,
"valueWidth": 0,
"convolutionWidth": 0,
"recurrentWidth": 0
}