hasktorch-gradually-typed-0.2.0.0: experimental project for hasktorch
Safe HaskellSafe-Inferred
LanguageHaskell2010

Torch.GraduallyTyped.NN.Transformer.T5.Small

Synopsis

Documentation

type T5SmallNumLayers = 6 Source #

T5-Small number of layers. 'num_layers = 6'

t5SmallNumLayers :: SNat T5SmallNumLayers Source #

T5-Small number of layers singleton.

type T5SmallHeadDim = 'Dim ('Name "*") ('Size 8) Source #

T5-Small number of attention heads. 'n_heads = 8'

type T5SmallHeadEmbedDim = 'Dim ('Name "*") ('Size 64) Source #

T5-Small head embedding dimension. 'd_kv = 64'

type T5SmallEmbedDim = 'Dim ('Name "*") ('Size 512) Source #

T5-Small embedding dimension. 'inner_dim = n_heads * d_kv = 512'

type T5SmallInputEmbedDim = 'Dim ('Name "*") ('Size 512) Source #

T5-Small model dimension. 'd_model = 512'

type T5SmallFFNDim = 'Dim ('Name "*") ('Size 2048) Source #

T5-Small feed-forward network dimension. 'd_ff = 2048'

type T5SmallVocabDim = 'Dim ('Name "*") ('Size 32128) Source #

T5-Small vocabulary dimension. 'vocab_size = 32128'

type T5Small (transformerHead :: TransformerHead) (gradient :: Gradient RequiresGradient) (device :: Device (DeviceType Nat)) (hasDropout :: HasDropout) = T5ModelF 'T5 transformerHead T5SmallNumLayers T5SmallNumLayers gradient device T5SmallHeadDim T5SmallHeadEmbedDim T5SmallEmbedDim T5SmallInputEmbedDim T5SmallFFNDim T5SmallVocabDim hasDropout Source #

T5-Small model.

t5SmallSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (T5Small transformerHead gradient device hasDropout) Source #

T5-Small model specification.

type ByT5SmallNumEncoderLayers = 12 Source #

ByT5-Small number of encoder layers. 'num_layers = 12'

byT5SmallNumEncoderLayers :: SNat ByT5SmallNumEncoderLayers Source #

ByT5-Small number of encoder layers singleton.

type ByT5SmallNumDecoderLayers = 4 Source #

ByT5-Small number of decoder layers. 'num_decoder_layers = 4'

byT5SmallNumDecoderLayers :: SNat ByT5SmallNumDecoderLayers Source #

ByT5-Small number of encoder layers singleton.

type ByT5SmallHeadDim = 'Dim ('Name "*") ('Size 6) Source #

ByT5-Small number of attention heads. 'n_heads = 6'

type ByT5SmallHeadEmbedDim = 'Dim ('Name "*") ('Size 64) Source #

ByT5-Small head embedding dimension. 'd_kv = 64'

type ByT5SmallEmbedDim = 'Dim ('Name "*") ('Size 384) Source #

ByT5-Small embedding dimension. 'inner_dim = n_heads * d_kv = 384'

type ByT5SmallInputEmbedDim = 'Dim ('Name "*") ('Size 1472) Source #

ByT5-Small model dimension. 'd_model = 1472'

type ByT5SmallFFNDim = 'Dim ('Name "*") ('Size 3584) Source #

T5-Small feed-forward network dimension. 'd_ff = 3584'

type ByT5SmallVocabDim = 'Dim ('Name "*") ('Size 384) Source #

T5-Small vocabulary dimension. 'vocab_size = 384'

byT5SmallSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (ByT5Small transformerHead gradient device hasDropout) Source #

ByT5-Small model specification.