| Safe Haskell | Safe-Inferred |
|---|---|
| Language | Haskell2010 |
Torch.GraduallyTyped.NN.Transformer.RoBERTa.Base
Synopsis
- type RoBERTaBaseNumLayers = 12
- robertaBaseNumLayers :: SNat RoBERTaBaseNumLayers
- type RoBERTaBaseHeadDim = 'Dim ('Name "*") ('Size 12)
- type RoBERTaBaseHeadEmbedDim = 'Dim ('Name "*") ('Size 64)
- type RoBERTaBaseEmbedDim = 'Dim ('Name "*") ('Size 768)
- type RoBERTaBaseInputEmbedDim = 'Dim ('Name "*") ('Size 768)
- type RoBERTaBaseFFNDim = 'Dim ('Name "*") ('Size 3072)
- type RoBERTaBaseVocabDim = 'Dim ('Name "*") ('Size 50265)
- robertaBaseVocabDim :: SDim RoBERTaBaseVocabDim
- type RoBERTaBaseTypeVocabDim = 'Dim ('Name "*") ('Size 1)
- type RoBERTaBase (transformerHead :: TransformerHead) (gradient :: Gradient RequiresGradient) (device :: Device (DeviceType Nat)) (hasDropout :: HasDropout) = RoBERTaModelF transformerHead RoBERTaBaseNumLayers gradient device RoBERTaBaseHeadDim RoBERTaBaseHeadEmbedDim RoBERTaBaseEmbedDim RoBERTaBaseInputEmbedDim RoBERTaBaseFFNDim RoBERTaBaseVocabDim RoBERTaBaseTypeVocabDim hasDropout
- robertaBaseSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (RoBERTaBase transformerHead gradient device hasDropout)
Documentation
type RoBERTaBaseNumLayers = 12 Source #
RoBERTa-Base number of layers. 'num_hidden_layers = 12'
robertaBaseNumLayers :: SNat RoBERTaBaseNumLayers Source #
RoBERTa-Base number of layers singleton.
type RoBERTaBaseHeadDim = 'Dim ('Name "*") ('Size 12) Source #
RoBERTa-Base number of attention heads. 'num_attention_heads = 12'
type RoBERTaBaseHeadEmbedDim = 'Dim ('Name "*") ('Size 64) Source #
RoBERTa-Base head embedding dimension. 'd_kv = 64'
type RoBERTaBaseEmbedDim = 'Dim ('Name "*") ('Size 768) Source #
RoBERTa-Base embedding dimension. 'hidden_size = n_heads * d_kv = 768'
type RoBERTaBaseInputEmbedDim = 'Dim ('Name "*") ('Size 768) Source #
RoBERTa-Base model dimension. 'hidden_size = 768'
type RoBERTaBaseFFNDim = 'Dim ('Name "*") ('Size 3072) Source #
RoBERTa-Base feed-forward network dimension. 'intermediate_size = 3072'
type RoBERTaBaseVocabDim = 'Dim ('Name "*") ('Size 50265) Source #
RoBERTa-Base vocabulary dimension. 'vocab_size = 50265'
robertaBaseVocabDim :: SDim RoBERTaBaseVocabDim Source #
RoBERTa-Base vocabulary dimension singleton.
type RoBERTaBaseTypeVocabDim = 'Dim ('Name "*") ('Size 1) Source #
RoBERTa-Base type vocabulary dimension. 'type_vocab_size = 1'
type RoBERTaBase (transformerHead :: TransformerHead) (gradient :: Gradient RequiresGradient) (device :: Device (DeviceType Nat)) (hasDropout :: HasDropout) = RoBERTaModelF transformerHead RoBERTaBaseNumLayers gradient device RoBERTaBaseHeadDim RoBERTaBaseHeadEmbedDim RoBERTaBaseEmbedDim RoBERTaBaseInputEmbedDim RoBERTaBaseFFNDim RoBERTaBaseVocabDim RoBERTaBaseTypeVocabDim hasDropout Source #
RoBERTa-Base model.
robertaBaseSpec :: STransformerHead transformerHead -> SGradient gradient -> SDevice device -> SHasDropout hasDropout -> ModelSpec (RoBERTaBase transformerHead gradient device hasDropout) Source #
RoBERTa-Base model specification.