| Safe Haskell | Safe-Inferred |
|---|---|
| Language | Haskell2010 |
Torch.GraduallyTyped.NN.Transformer.BART.Large
Synopsis
- type BARTLargeNumLayers = 12
- bartLargeNumLayers :: SNat BARTLargeNumLayers
- type BARTLargeHeadDim = 'Dim ('Name "*") ('Size 16)
- bartLargeHeadDim :: SDim BARTLargeHeadDim
- type BARTLargeHeadEmbedDim = 'Dim ('Name "*") ('Size 64)
- bartLargeHeadEmbedDim :: SDim BARTLargeHeadEmbedDim
- type BARTLargeEmbedDim = 'Dim ('Name "*") ('Size 1024)
- bartLargeEmbedDim :: SDim BARTLargeEmbedDim
- type BARTLargeInputEmbedDim = 'Dim ('Name "*") ('Size 1024)
- bartLargeInputEmbedDim :: SDim BARTLargeInputEmbedDim
- type BARTLargeFFNDim = 'Dim ('Name "*") ('Size 4096)
- bartLargeFFNDim :: SDim BARTLargeFFNDim
- type BARTLargeVocabDim = 'Dim ('Name "*") ('Size 50265)
- bartLargeVocabDim :: SDim BARTLargeVocabDim
- type BARTLarge (transformerHead :: TransformerHead) (gradient :: Gradient RequiresGradient) (device :: Device (DeviceType Nat)) (hasDropout :: HasDropout) = BARTModelF transformerHead BARTLargeNumLayers gradient device BARTLargeHeadDim BARTLargeHeadEmbedDim BARTLargeEmbedDim BARTLargeInputEmbedDim BARTLargeFFNDim BARTLargeVocabDim hasDropout
- bartLargeSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (BARTLarge transformerHead gradient device hasDropout)
Documentation
type BARTLargeNumLayers = 12 Source #
BART-Large number of layers. 'encoder_layers = 12' 'decoder_layers = 12'
bartLargeNumLayers :: SNat BARTLargeNumLayers Source #
BART-Large number of layers singleton.
type BARTLargeHeadDim = 'Dim ('Name "*") ('Size 16) Source #
BART-Large number of attention heads. 'encoder_attention_heads = 16' 'decoder_attention_heads = 16'
bartLargeHeadDim :: SDim BARTLargeHeadDim Source #
BART-Large number of attention heads singleton.
type BARTLargeHeadEmbedDim = 'Dim ('Name "*") ('Size 64) Source #
BART-Large head embedding dimension. 'd_kv = 64'
bartLargeHeadEmbedDim :: SDim BARTLargeHeadEmbedDim Source #
BART-Large head embedding dimension singleton.
type BARTLargeEmbedDim = 'Dim ('Name "*") ('Size 1024) Source #
BART-Large embedding dimension. 'hidden_size = n_heads * d_kv = 1024'
bartLargeEmbedDim :: SDim BARTLargeEmbedDim Source #
BART-Large embedding dimension singleton.
type BARTLargeInputEmbedDim = 'Dim ('Name "*") ('Size 1024) Source #
BART-Large model dimension. 'd_model = 1024'
bartLargeInputEmbedDim :: SDim BARTLargeInputEmbedDim Source #
BART-Large model dimension singleton.
type BARTLargeFFNDim = 'Dim ('Name "*") ('Size 4096) Source #
BART-Large feed-forward network dimension. 'encoder_ffn_dim = 4096' 'decoder_ffn_dim = 4096'
bartLargeFFNDim :: SDim BARTLargeFFNDim Source #
BART-Large feed-forward network dimension singleton.
type BARTLargeVocabDim = 'Dim ('Name "*") ('Size 50265) Source #
BART-Large vocabulary dimension. 'vocab_size = 50265'
bartLargeVocabDim :: SDim BARTLargeVocabDim Source #
BART-Large vocabulary dimension singleton.
type BARTLarge (transformerHead :: TransformerHead) (gradient :: Gradient RequiresGradient) (device :: Device (DeviceType Nat)) (hasDropout :: HasDropout) = BARTModelF transformerHead BARTLargeNumLayers gradient device BARTLargeHeadDim BARTLargeHeadEmbedDim BARTLargeEmbedDim BARTLargeInputEmbedDim BARTLargeFFNDim BARTLargeVocabDim hasDropout Source #
BART-Large model.
bartLargeSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (BARTLarge transformerHead gradient device hasDropout) Source #
BART-Large model specification.