| | |
|---|
| | | att_layer_num: int = 6, |
|---|
| | | kernel_size: int = 21, |
|---|
| | | sanm_shfit: int = 0, |
|---|
| | | lora_list: List[str] = None, |
|---|
| | | lora_rank: int = 8, |
|---|
| | | lora_alpha: int = 16, |
|---|
| | | lora_dropout: float = 0.1, |
|---|
| | | tf2torch_tensor_name_prefix_torch: str = "decoder", |
|---|
| | | tf2torch_tensor_name_prefix_tf: str = "seq2seq/decoder", |
|---|
| | | ): |
|---|
| | |
|---|
| | | attention_dim, self_attention_dropout_rate, kernel_size, sanm_shfit=sanm_shfit |
|---|
| | | ), |
|---|
| | | MultiHeadedAttentionCrossAtt( |
|---|
| | | attention_heads, attention_dim, src_attention_dropout_rate |
|---|
| | | attention_heads, attention_dim, src_attention_dropout_rate, lora_list, lora_rank, lora_alpha, lora_dropout |
|---|
| | | ), |
|---|
| | | PositionwiseFeedForwardDecoderSANM(attention_dim, linear_units, dropout_rate), |
|---|
| | | dropout_rate, |
|---|
