Package ai.djl.nn.transformer
Contains blocks for transformer models.
-
Class Summary Class Description BertBlock Implements the core bert model (without next sentence and masked language task) of bert.BertBlock.Builder BertMaskedLanguageModelBlock Block for the bert masked language task.BertMaskedLanguageModelLoss The loss for the bert masked language model task.BertNextSentenceBlock Block to perform the Bert next-sentence-prediction task.BertNextSentenceLoss Calculates the loss for the next sentence prediction task.BertPretrainingBlock Creates a block that performs all bert pretraining tasks (next sentence and masked language).BertPretrainingLoss Loss that combines the next sentence and masked language losses of bert pretraining.IdEmbedding An Embedding from integer ids to float vectors.IdEmbedding.Builder The Builder to construct anIdEmbedding
type ofBlock
.MissingOps Operators missing from NDArray that are necessary to implement Bert pretraining.PointwiseFeedForwardBlock Fully connected Feed-Forward network, only applied to the last dimension of the input.ScaledDotProductAttentionBlock A Block implementing scaled product attention according to Vaswani et.ScaledDotProductAttentionBlock.Builder A builder forScaledDotProductAttentionBlock
s.TransformerEncoderBlock Self-Attention based transformer encoder block.