mlx-bun library API
Documentation
Preparing search index...
@mlx-bun/inference
@mlx-bun/inference/kernels/attention/paged
PagedAttentionStorage
Interface PagedAttentionStorage
Borrowed numerical storage. Allocation, appends, and eviction belong to state.
interface
PagedAttentionStorage
{
blockSize
:
number
;
dtype
:
Dtype
;
headDim
:
number
;
keyBiases
?:
MlxArray
|
null
;
keys
:
MlxArray
;
keyScales
?:
MlxArray
|
null
;
quantization
?:
{
bits
:
number
;
groupSize
:
number
}
|
null
;
valueBiases
?:
MlxArray
|
null
;
values
:
MlxArray
;
valueScales
?:
MlxArray
|
null
;
vHeadDim
:
number
;
}
Index
Properties
block
Size
dtype
head
Dim
key
Biases?
keys
key
Scales?
quantization?
value
Biases?
values
value
Scales?
v
Head
Dim
Properties
Readonly
block
Size
blockSize
:
number
Readonly
dtype
dtype
:
Dtype
Readonly
head
Dim
headDim
:
number
Optional
Readonly
key
Biases
keyBiases
?:
MlxArray
|
null
Readonly
keys
keys
:
MlxArray
Optional
Readonly
key
Scales
keyScales
?:
MlxArray
|
null
Optional
Readonly
quantization
quantization
?:
{
bits
:
number
;
groupSize
:
number
}
|
null
Optional
Readonly
value
Biases
valueBiases
?:
MlxArray
|
null
Readonly
values
values
:
MlxArray
Optional
Readonly
value
Scales
valueScales
?:
MlxArray
|
null
Readonly
v
Head
Dim
vHeadDim
:
number
Settings
Member Visibility
Protected
Inherited
External
Theme
OS
Light
Dark
On This Page
Properties
block
Size
dtype
head
Dim
key
Biases
keys
key
Scales
quantization
value
Biases
values
value
Scales
v
Head
Dim
Documentation
mlx-bun library API
Loading...
Borrowed numerical storage. Allocation, appends, and eviction belong to state.