mlx-bun library API
Documentation
Preparing search index...
@mlx-bun/inference
@mlx-bun/inference/layers
Module @mlx-bun/inference/layers
Classes
Affine3MmaLinear
Affine3RowsLinear
Affine4MmaLinear
Affine4RowsLinear
AffineRowLinear
AffineSplitMmaLinear
AneAffineSplit
AneTrellisExpandedMlpSplit
AneTrellisExpandedMlpSplitK3i
AneTrellisExpandReduce
AneTrellisExpandScatter
AneTrellisMlpSplit
AneTrellisMlpSplitK3i
CodebookDownK3iRow
CodebookDownRow
CodebookGateUpRow
CodebookMmaDown
CodebookMmaDownK3i
CodebookMmaGateUp
CodebookSplitGateUpRow
CompiledGatedRMSNorm
CompiledGeglu
CompiledGegluMLP
CompiledResidualRMSNorm
CompiledSiluRMSHeads
DenseEmbedding
DenseLinear
DequantizedLinear
DequantizedMatmulLinear
ExpandedTrellisLinear
FactoredDownK3iRow
FactoredDownK3iRows
FactoredDownRow
FactoredDownRows
FactoredGateUpRow
FactoredGateUpRows
FactoredMixedGateUp
GegluMLP
LayerNorm
MmaDown
MmaDownK3i
MmaDownWide
MmaDownWideK3i
MmaGateUp
MmaGateUpWide
NoRope
PlainGeglu
PlainGegluMLP
QuantizedEmbedding
QuantizedLinear
QuantizedLinearPlain
QuantizedLinearRows
QuantizedSwitchLinear
ReduceGateUpRow
ReduceGateUpRows
ReduceMixedGateUp
ReduceSplitMixedGateUp
RMSNorm
ScatterDownK3iRow
ScatterDownK3iRows
ScatterDownRow
ScatterDownRows
TrellisExpandReducePrompt
TrellisExpandScatterPrompt
TrellisFusedMLP
TrellisLinear
TrellisReducePrompt
TrellisScatterPrompt
TrellisSplitKPrompt
TrellisTiledPrompt
TrellisUnfusedMLP
TrellisWidePrompt
UniversalRope
Interfaces
QuantizedLinearState
Rope2dTables
RowProjection
Type Aliases
AneGpuMlpRows
AneGpuRows
AneMlpSplit
AnyEmbedding
AnyLinear
AnyNorm
RopeScalingConfig
Variables
DEQUANTIZED_MIN_ROWS
FINFO_MIN
FUSED_N_CHUNK
Functions
aneAffineSplitProgram
aneChannels
aneMlpSplitProgram
applyInterleavedRope
applyRope2d
bidirMask
buildMropePositions
compiledGeglu
compiledSwiglu
createCausalMask
dequantizedMinRows
disposing
foldedQuantizedSdpa
foldedQuantizedSdpaEligible
foldedSdpa
foldedSdpaEligible
fusedGateUpEligible
fusedGateUpSwiglu
fusedSdpaConfigSupported
geglu
initializeRope
interleavedSelector
loadEmbedding
loadLayerNorm
loadLinear
loadRmsNorm
mropeInvFreq
mropePositionIds
phi3SuRope
quantizedAppendAttention
quantizedSdpaGroupedHeads
quantizedSdpaTiled
quantizedSdpaUnfused
qwenRopeIndex
rope2dTables
trellisGeometry
trellisRows
trellisStoredRows
References
LoraDelta
→
LoraDelta
LoraOperation
→
LoraOperation
LoraState
→
LoraState
MixedGateUpTail
→
MixedGateUpTail
MropeForwardState
→
MropeForwardState
MropeRequestState
→
MropeRequestState
TRELLIS_MATVEC_MAX_M
→
TRELLIS_MATVEC_MAX_M
TrellisGeometry
→
TrellisGeometry
Settings
Member Visibility
Protected
Inherited
External
Theme
OS
Light
Dark
On This Page
Classes
Affine3
Mma
Linear
Affine3
Rows
Linear
Affine4
Mma
Linear
Affine4
Rows
Linear
Affine
Row
Linear
Affine
Split
Mma
Linear
Ane
Affine
Split
Ane
Trellis
Expanded
Mlp
Split
Ane
Trellis
Expanded
Mlp
Split
K3i
Ane
Trellis
Expand
Reduce
Ane
Trellis
Expand
Scatter
Ane
Trellis
Mlp
Split
Ane
Trellis
Mlp
Split
K3i
Codebook
Down
K3i
Row
Codebook
Down
Row
Codebook
Gate
Up
Row
Codebook
Mma
Down
Codebook
Mma
Down
K3i
Codebook
Mma
Gate
Up
Codebook
Split
Gate
Up
Row
Compiled
Gated
RMS
Norm
Compiled
Geglu
Compiled
Geglu
MLP
Compiled
Residual
RMS
Norm
Compiled
Silu
RMS
Heads
Dense
Embedding
Dense
Linear
Dequantized
Linear
Dequantized
Matmul
Linear
Expanded
Trellis
Linear
Factored
Down
K3i
Row
Factored
Down
K3i
Rows
Factored
Down
Row
Factored
Down
Rows
Factored
Gate
Up
Row
Factored
Gate
Up
Rows
Factored
Mixed
Gate
Up
Geglu
MLP
Layer
Norm
Mma
Down
Mma
Down
K3i
Mma
Down
Wide
Mma
Down
Wide
K3i
Mma
Gate
Up
Mma
Gate
Up
Wide
No
Rope
Plain
Geglu
Plain
Geglu
MLP
Quantized
Embedding
Quantized
Linear
Quantized
Linear
Plain
Quantized
Linear
Rows
Quantized
Switch
Linear
Reduce
Gate
Up
Row
Reduce
Gate
Up
Rows
Reduce
Mixed
Gate
Up
Reduce
Split
Mixed
Gate
Up
RMS
Norm
Scatter
Down
K3i
Row
Scatter
Down
K3i
Rows
Scatter
Down
Row
Scatter
Down
Rows
Trellis
Expand
Reduce
Prompt
Trellis
Expand
Scatter
Prompt
Trellis
Fused
MLP
Trellis
Linear
Trellis
Reduce
Prompt
Trellis
Scatter
Prompt
Trellis
Split
K
Prompt
Trellis
Tiled
Prompt
Trellis
Unfused
MLP
Trellis
Wide
Prompt
Universal
Rope
Interfaces
Quantized
Linear
State
Rope2d
Tables
Row
Projection
Type Aliases
Ane
Gpu
Mlp
Rows
Ane
Gpu
Rows
Ane
Mlp
Split
Any
Embedding
Any
Linear
Any
Norm
Rope
Scaling
Config
Variables
DEQUANTIZED_
MIN_
ROWS
FINFO_
MIN
FUSED_
N_
CHUNK
Functions
ane
Affine
Split
Program
ane
Channels
ane
Mlp
Split
Program
apply
Interleaved
Rope
apply
Rope2d
bidir
Mask
build
Mrope
Positions
compiled
Geglu
compiled
Swiglu
create
Causal
Mask
dequantized
Min
Rows
disposing
folded
Quantized
Sdpa
folded
Quantized
Sdpa
Eligible
folded
Sdpa
folded
Sdpa
Eligible
fused
Gate
Up
Eligible
fused
Gate
Up
Swiglu
fused
Sdpa
Config
Supported
geglu
initialize
Rope
interleaved
Selector
load
Embedding
load
Layer
Norm
load
Linear
load
Rms
Norm
mrope
Inv
Freq
mrope
Position
Ids
phi3
Su
Rope
quantized
Append
Attention
quantized
Sdpa
Grouped
Heads
quantized
Sdpa
Tiled
quantized
Sdpa
Unfused
qwen
Rope
Index
rope2d
Tables
trellis
Geometry
trellis
Rows
trellis
Stored
Rows
References
Lora
Delta
Lora
Operation
Lora
State
Mixed
Gate
Up
Tail
Mrope
Forward
State
Mrope
Request
State
TRELLIS_
MATVEC_
MAX_
M
Trellis
Geometry
Documentation
mlx-bun library API
Loading...