Interface UNetModelConfig

Hierarchy

  • UNetModelConfig

Properties

attentionResolutions?: number[]

a collection of downsample rates at which attention is applied. For example, if this contains 4, then at 4x downsampling, attention is applied.

channelMult?: number[]

channel multiples per down/up sampling stage

contextDim?: number
convResample?: boolean

if true use learnable convolutional upsampling/downsampling

dims?: number

determines whether to use 1D, 2D, or 3D convolutions

dropout?: number

the dropout probability

dtype?: Dtype
inChannels: number

channels of input tensor

modelChannels: number

base channels in model

numClasses?: number

if specified, then this model will be class-conditioned with numClasses classes

numHeadChannels?: number

if specified, ignore numHeads and instead use this number of channels in each attention head

numHeads?: number

the number of attention heads in each attention layer

numHeadsUpSample?: number
numResBlocks?: number

number of residual blocks per down/up sampling stage

outChannels: number

channels of output tensor

resblockUpdown?: boolean

use residual blocks for up/down sampling

transformerDepth?: number
useCheckpoint?: boolean
useNewAttentionOrder?: boolean
useScaleShiftNorm?: boolean

use a FiLM-like conditioning mechanism

useSpatialTransformer?: boolean

Generated using TypeDoc